HUMAN
INTELLIGENCE
EARL HUNT
Human Intelligence
This book is a comprehensive survey of our scientific knowledge about human intelligence,
written by a researcher who has spent more than thirty years studying the field. It takes a
nonideological view of a topic in which, too often, writings are dominated by a single theory
or social viewpoint. The book discusses the conceptual status of intelligence as a collection
of cognitive skills that include, but also go beyond, those skills evaluated by conventional
tests; intelligence tests and their analysis; contemporary theories of intelligence; biological
and social causes of intelligence; the importance of intelligence in social, industrial, and
educational spheres; the role of intelligence in determining success in life, both inside and
outside educational settings; and the nature and causes of variations in intelligence across
age, gender, and racial and ethnic groups.
Earl Hunt is Professor Emeritus at the University of Washington, where he has been a
faculty member since 1966. He has also taught at Yale; the University of California, Los
Angeles; and the University of Sydney, Australia. His other books include Concept Learning
(1962), Experiments in Induction (1966), Artificial Intelligence (1975), Will We Be Smart Enough?
(1995], Thoughts on Thought (2002), and The Mathematics of Behavior (2007). He has received
the International Society for Intelligence Research’s Lifetime Achievement Award for his
contributions to the study of intelligence and has been named the 2011 recipient of the
Association for Psychological Science’s Cattell Award for lifetime contributions to applied
psychological research.
Human Intelligence
EARL HUNT
University of Washington
Cambridge
UNIVERSITY PRESS
CAMBRIDGE UNIVERSITY PRESS
Cambridge, New York, Melbourne, Madrid, Cape Town, Singapore,
Sao Paulo, Delhi, Dubai, Tokyo, Mexico City
Cambridge University Press
32 Avenue of the Americas, New York, NY 10013-2473, USA
www .Cambridge . org
Information on this title: www.cambridge.0rg/9780521707817
© Earl Hunt 2011
This publication is in copyright. Subject to statutory exception
and to the provisions of relevant collective licensing agreements,
no reproduction of any part may take place without the written
permission of Cambridge University Press.
First published 2011
Printed in the United States of America
A catalog record for this publication is available from the British Library.
Library of Congress Cataloging in Publication data
Hunt, Earl B.
Human intelligence / Earl Hunt,
p. cm.
Includes bibliographical references and index.
ISBN 978-0-521-88162-3 (hardback) - ISBN 978-0-521-70781-7 (pbk.)
1. Intellect. I. Title.
BF431.H814 2010
i53*9-dc22 2010038696
ISBN 978-0-521-88162-3 Hardback
ISBN 978-0-521-70781-7 Paperback
Cambridge University Press has no responsibility for the persistence or accuracy of URLs for external or third-party
Internet Web sites referred to in this publication and does not guarantee that any content on such Web sites is, or
will remain, accurate or appropriate.
Contents
List of Figures
page vii
List of Tables
xi
Introductory Remarks
xiii
1 The Issue of Intelligence
l
2 The Tests
31
3 On Theory
64
4 Psychometric Theories
79
5 Taking Intelligence Beyond Psychometrics
111
6 The Mechanics of Intelligence
140
7 Intelligence and the Brain
172
8 The Genetic Basis of Intelligence
203
9 Environmental Effects on Intelligence
257
10 What Use Is Intelligence?
3U
1 1 The Demography of Intelligence
356
1 2 Summary and Prospectus
448
References
453
Author Index
485
Subject Index
494
List of Figures
1.1. The Bell Curve page 9
1.2. Survival rates for individuals in the upper and lower quartiles of the
intelligence distribution, shown separately for men and women 15
1.3. The Hunt and Carlson model of the development of intelligence 19
1.4. Intelligence and socially relevant behavior 23
1.5. The concept of reaction range 25
1.6. Interactions between reaction ranges and environmental quality 26
1.7. The system of variables influencing the development of intelligence during
the school years 29
2.1. The four major lobes of the cerebral cortex 37
2.2. Two progressive matrix items 47
2.3. An example of a visual-spatial reasoning problem 49
2.4. College/university graduation rates as a function of a student’s SAT scores 59
3.1. Genetic and environmental influences on intelligence 67
3.2. Direct and indirect influences on reading 69
3.3. Levels of theories of intelligence 71
4.1 A hypothetical example illustrating factor analysis 81
4.2. The first and second factors in the hypothetical example 82
4.3. Correlated factors extracted from the hypothetical example 83
4.4. A model for the confirmatory factor analysis example 85
4.5. A general factor model for Spearman’s data 91
4.6. Spearman’s general and specific factor model applied to a hypothetical
test battery 92
4.7. A hierarchical model applied to the hypothetical battery 93
4.8. The structure of the second-stratum and third-stratum abilities in
the WJ-III test 100
4.9. The railroad passing problem 105
vii
viii
LIST OF FIGURES
4.10. Vernon’s structure of intelligence model 106
4.11. The structure of Johnson and Bouchard’s VPR model 107
5.1. The relation between underlying traits, social constraints, test scores,
and socially relevant behaviors 114
6.1. The blackboard model of cognition 144
6.2. The inspection time paradigm 152
6.3. A progressive matrix item of intermediate difficulty 161
6.4. A visual closure task 169
6.5. A rotation problem 169
7.1. A sketch of the human brain, seen from the left 174
7.2. The cingulate gyrus and the limbic system 175
7.3. A sketch of areas of the brain that show activity during a variety
of tasks involving working memory 191
7.4. Broca’s and Wernicke’s regions on the left side of the brain 193
7.5. Areas that have shown activation during selected reading tasks 194
7.6. The dorsal and ventral visual streams 194
7.7. Regions of the brain activated within a single individual when that person
was instructed to solve a sentence comprehension problem using either a
verbal or a visual imagery strategy 197
7.8. The relation between brain systems and broad (second-stratum)
psychometric abilities 199
7.9. Hypothetical relationships between brain systems, narrowly defined
information-processing functions, working memory, and general reasoning
ability (g) 201
8.1. A simple illustration of Mendelian genetics 206
8.2. The effect of evolutionary selection on the distribution of genotypes 209
8.3. The distribution of phenotypes with independent environmental effects 211
8.4. Probabilities for IQ potential (ordinate) as a function of the value of the
genotypic potential 214
8.5. The relationship between genetic and environmental contributions in
a single individual 221
8.6. A model of the relation between genetic and environmental effects
in two individuals, i and i' 222
8.7. The ACE model 226
8.8. Percentage of the variance in intelligence test score accounted for by additive
genetic influence, shared environmental influence, and nonshared
environmental influence 233
8.9. A schematic of genetic correlations involving the P and R factors in
the VPR model of intelligence 237
8.10. The relative proportions of variance associated with genetic and
environmental variance in the factors defined for the g-VPR model 238
8.11. Heritability estimates for three academic skills 240
8.12. The reshuffling of genetic material in meiosis 246
9.1. The design of a hypothetical experiment demonstrating how the development
of cognition may depend on environments that encourage exploration 259
9.2. Scores on the Army Alpha Test obtained by World War I soldiers, World
War II soldiers, and literate World War I soldiers 263
9.3. The increase in three dimensions of intelligence over cohorts, using the 1903
birth cohort as a base 264
9.4. Progressive matrix test scores for eighteen-year-old Danish men registering
for military enlistment, 1958-98 267
LIST OF FIGURES ix
9.5. The fraction of variance in WISC scores predictable from measures of SES
and home environment 288
9.6. Mean IQ scores obtained by the participants in the ABCDerian project and
in a randomly chosen control group 291
9.7. Mean intelligence estimates of young children bom to mothers who were
in the NLSY79 survey 293
9.8. The relation between number of children and maternal intelligence score
(AFQT percentile score] in the NLSY79 data set 294
9.9. The birth order effect appears in families where a child becomes the oldest
due to the death of an elder sibling 295
9.10. The path diagram for a study showing the influence of education
on intelligence 299
9.11. A problem similar to the problems presented in K-2 mathematics textbooks 300
9.12. Abacus training improves intelligence test scores 303
9.13. A cartoon rendition of Posner and Rothbart’s attention training task 306
10.x. Percentage of white young adults in the NLSY79 survey who did not
complete high school, plotted as a function of their percentile scores
on the AFQT 320
10.2. The distribution of RSPM scores across grades in a representative sample of
500 Icelandic schoolchildren 321
10.3. Test accuracy and rejection rate interact to produce quality acceptances 325
10.4. Median weekly earnings in 2008 as a function of level of education 326
10.5. Correlations between predictors and criterion measures in the U.S. Army
study of enlisted performance 330
10.6. The correlations between measures of job performance, measures of general
intelligence, and a variety of other assessment measures 331
10.7. Predicting square root income (SQRT INCOME] and occupational prestige
(SEI] in 1995 from educational attainment (in years] and AFQT score
obtained in 1980 337
10.8. Correlations and multiple correlations between income (top) and
occupational status (bottom) and general intelligence and educational level 339
10.9. Recommended GATE cut points for four occupations 340
10.10. The ninetieth, fiftieth (median), and tenth decile of incomes in various
occupations, plotted as a function of the imputed intelligence demands
for the occupation 341
10.11. Educational attainment of participants in the SMPY, by age thirty-three 347
10.12. The SMPY participants compared to graduate students of the same age 348
10.13. Attrition rates during basic training (percentages) for Project 100,000
participants and for the service as a whole, broken down by military service 352
10.14. The advancement in rank of Project 100,000 personnel 353
11.1. An illustration of the overlap between distributions 362
11.2. Broad Cognitive Ability (BCA) factor scores as a function of age 368
11.3. Age and change scores for Gf and Gc 369
11.4. A stylized picture of the age trends for different broad second-level traits
in the Gc-Gf model of intelligence 370
11.5. Cumulative age changes from age twenty-five for three different aspects
of intelligence 371
11.6. Reaction times lengthen with age 372
11.7. Changes in the level of a generalized cognitive trait (g) over the second half
of the life span 373
11.8. Scores on a test of basic living skills, measured on two different occasions 374
X
LIST OF FIGURES
11.9. The distribution of CAT mean scores in a British sample of more than
300,000 eleven-year-old schoolchildren 383
11.10. The effect of small differences in means and variances on the distribution
of scores at various points on the d scale 385
11.11. Male-female differences in d unit for selected subtests of the Differential
Aptitude Battery 388
11.12. Johnson and Bouchard’s model of general intelligence plus two bipolar factors
(verbal-rotational and focus-diffuse) 389
11.13. Illustrations of two spatial-visual reasoning tasks on which men tend to
outperform women 389
11.14. Detecting details in a complex picture 390
11.15. Male-female differences on various educational tests of reading and
mathematics 392
11.16. The difference between male and female scores on three of the 2003 PISA
examinations 393
11.17. Mean scores of fifteen-year-old students in selected countries on the PISA
2003 tests of mathematics, science, and general problem solving 394
11.18. Mean SAT mathematics scores for men and women, from 1988 to 2008 395
11.19. The relation between SAT mathematics scores, grades (A, B, C, D, E), and
gender in three levels of college courses 396
11.20. The percentages of women among the recipients of doctoral degrees, 1977
and 2007 398
11.21. Raven Progressive Matrices scores in a school district in the western United
States, as a function of age and racial/ethnic group 413
11.22. The median number of Raven’s Standard Progressive Matrices problems
solved by Native American, Mestizo, and White children in Mexico 413
11.23. Median number of problems solved as a function of age and ethnicity 414
11.24. The difference between White seventeen-year-olds and African American
and Hispanic seventeen-year-olds on the NAEP mathematics test 418
11.25. White, African American, and Hispanic mean scores on the SAT 419
11.26. Difference in deviation score units between Black and White students’ scores
on achievement tests taken in the second and twelfth grades 419
11.27. SAT scores for three racial/ethnic groups for the period 1997-2007 422
11.28. The effect sizes for White-Black contrasts on selected variables shown
to predict the Black- White contrast on tests of cognitive skills 429
List of Tables
1.1. The distributions of standard and IQ scores page 8
2.1. The subscales of the Wechsler Adult Intelligence Scale-IV 35
2.2. The subtests of the Kaufman Adult Intelligence Test 36
2.3. The subtests of the Cognitive Assessment System 38
2.4. The Otis-Lennon test of School Abilities 40
2.5. The scales and subtests of the Cognitive Abilities Test-3 41
2.6. The subtests of the SAT-I 4 2
2.7. The subtests of the Armed Services Vocational Aptitude Battery (ASVAB) 43
2.8. Sample questions illustrating the Wonderlic Personnel Test, revised edition 45
4.1. Variance-covariance matrices in the confirmatory factor analysis sample 85
4.2. The correlation matrix for Spearman's data 90
4.3. Abilities identified in the Cattell-Horn-Carroll three-stratum theory 101
4.4. The g loadings of the second-stratum factors on the WJ-III test 103
5.1. Examples of tasks used by Sternberg and his colleagues to assess practical
intelligence 123
5.2. Selected data on the relation between knowledge and intelligence test
scores and age 133
6.1. Correlations between different visual-spatial abilities 166
6.2. Loadings of visual-spatial and imagery first-order factors on a general
visual- spatial reasoning factor 167
8.1. Genotypes and associated IQ potentials for the (artificially simple) case of
two genes affecting intelligence 214
8.2. Correlations and heritability estimates for pairs of individuals of various
degrees of relationships 227
9.1. Levels of blood lead concentration in children, together with
recommended actions 282
xi
xii LIST OF TABLES
9.2. Mean IQ scores of adopted children as pre-schoolers and as adolescents,
compared to scores of biological children of the adopting family 285
9.3. Mean WISC-R scores as a function of SES of birth and adopting parents 286
10.1. Correlations between tests used for college/university selection, the ASVAB
general factor or AFQT, and Raven's Advanced Progressive Matrices 322
10.2. Odds ratios comparing probability of graduation for the top and bottom
halves of admission test scores in an entering population of post-graduate
students 328
10.3. Cognitive skills assessed in USAAF aviation cadets, shown by occupation
followed after WW II 335
10.4. Correlations between AFQT scores obtained at ages sixteen to eighteen and
measures of social outcomes when the participants were in their early thirties 336
10.5. Gottfredson's examples of intelligence levels (on IQ scale) associated with
different occupations and techniques of information processing 338
10.6. Percentages of men and women in the Terman study who attained various
levels of education 344
10.7. Family income distribution of Terman study participants in the 1950s,
compared to “urban white families” at that time 345
10.8. Attrition rates during basic training (percentages) for Project 100,000
participants and for the service as a whole, broken down by military branch 351
11.1. Ages at peak value and rate of decline for selected factors evaluated by
the WJ-R test 369
11.2. Selected scores and male-female comparison for brother-sister pairs in
the NLSY79 data set 379
11.3. Male-female standard deviation unit ( d ) scores for effect size for different
aspects of intelligence 387
11.4. Values of (White mean-African American mean) and (White mean-Latino
mean) in standard deviation units for a variety of cognitive tests 412
11.5. Correlations between IQ estimates and selected indices of national
well-being 438
11.6. Squared correlations and partial correlations between target variables
and IQ scores and education variables 441
Introductory Remarks
After spending some thirty years studying
human intelligence, I decided to write a
book. Why? It must be that I have some¬
thing new to say.
Well, I do and I do not. Virtually every¬
thing that could be said about human intel¬
ligence has been said. It is or isn’t impor¬
tant, it is or isn't evaluated by the tests, it
is or isn't genetically based, and on and on.
This book does not say anything that has
not already been said. It could not. What
it does do is attempt to bring up to date
the information on the various conflicting
views. My contribution is to moderate these
views, because I do not think that any of
the extreme statements that have been made
can be supported.
There is a saying that has been traced back
to the days of classic Greece;
The fox knows many little things; the hedge¬
hog knows one big thing.
It is not entirely clear what that means,
but some philosophers have interpreted it as
saying that some people summarize issues
with detailed, nuanced views, while others
make bold, simple statements. I am a fox. I
think the field of human intelligence has had
far too many hedgehogs. There are major
individual differences in cognitive power;
these differences have important implica¬
tions for human behavior; they do not have
a single cause, nor do they ever act outside
of the context of the current problem. We
need to understand intelligence in its full
complexity.
Intellectual foxes have a problem. They
are more likely to be right than intellectual
hedgehogs (there is actually data on this!),
but they are less likely to be believed (there
is data on this, too). Nevertheless, being a
fox, there is nothing I can do but try to locate
the burrows of as many intellectual hedge¬
hogs as I can, and try to dig them out. It is my
nature, and that is what I have tried to do.
Complete intellectual objectivity is impossi¬
ble to achieve. I have tried to present as fair
a picture as I can of a much-studied, much-
debated topic. The result is a book that may
sometimes be difficult to read, but I hope
that it is a comprehensive presentation.
Any effort of this sort is impossible unless
you receive support. My first and greatest
xiii
xiv
INTRODUCTORY REMARKS
debt is to my wife, Mary Lou Hunt, who
has put up with years of papers scattered all
over the house, a somewhat grumpy hus¬
band, and mutterings as I uncovered the
tracks of one or another of those intellec¬
tual hedgehogs.
My second debt is to Cambridge Univer¬
sity Press, which put up with my being late,
late, late, but let me persevere. I also owe
a special debt to Jeanie Lee, for her sub¬
stantial assistance in ensuring that permis¬
sions for reproduction were obtained. Too
many books on intelligence wave words at
the reader about what the data said. Thanks
in no small part to Ms. Lee’s assistance, this
book will often let the reader see what actu¬
ally was found.
I owe favors to colleagues around the
world who were willing to read pre¬
publication versions. Special thanks go to
Tom Bouchard Jr., who engaged me in lively
e-mail discussions over virtually every chap¬
ter; to two of my sons, Alan and Steven, who
discussed and commented on different top¬
ics (very different - Alan’s a biophysicist and
Steve an industrial-organizational psycholo¬
gist); to Wendy Johnson of the University
of Edinburgh for her comments on genet¬
ics; and to Diane Halpern, for comments
on the introductory chapters. Naturally, I
am responsible for everything in the final
product!
And I suppose I owe an apology to all
those authors whose works I should have
read but did not. All 1 can say is that
life is short and there are an awful lot of
you.
Bellevue, Washington
April 2010
Human Intelligence
CHAPTER 1
The Issue of Intelligence
So it is that gods do not give all men the
gifts of grace . . . neither good looks nor
intelligence nor eloquence.
Homer, The Odyssey
There’s many a man has more hair than wit.
Shakespeare, Comedy of Errors,
act 2, scene 3
1.1. The Idea of Intelligence
Homer and Shakespeare lived in very dif¬
ferent times, more than two thousand years
apart, but they both captured the same idea;
we are not all equally intelligent. I suspect
that anyone who has failed to notice this
is somewhat out of touch with the species.
However, we cannot simply sort people into
the “intelligent’' and the “not-so-intelligent.”
Homer observed that few people have great
gifts. Shakespeare, more pithily, observed
that all too many of us do not do terribly
well at problem solving. Most of us, though,
fall in between Homer’s desire for eloquence
and Shakespeare’s worry about lack of wit.
In this book I will talk about the nature
of intelligence, its causes, who has it, and
how it is used. I will do so without the elo¬
quence of Homer and Shakespeare. I will
take a scientific view. Modern psychology
has a great deal to say about intelligence, and
somehow a great deal that has been said has
been seriously misunderstood. The popular
media sometimes report that the psycholo¬
gists who study intelligence say almost the
opposite of what the psychologists actually
said.1
There is a reason for this. The study
of intelligence is not an isolated academic
topic; our intelligence has social conse¬
quences. We want our leaders to be intel¬
ligent, and exhibit concern if we think they
are not. There were politically motivated
attacks on the intelligence of Presidents
Lincoln, Truman, Harding, and Ford. Seri¬
ous concerns about mental competence
1 See Tannenbaum (1996) for a discussion of this issue
and references to earlier discussions of the topic.
Gottfredson (2005] provides a spirited discussion
of how failing to consider the implications of psy¬
chological research on intelligence can be costly to
society.
1
2
HUMAN INTELLIGENCE
were raised about Wilson and Eisenhower,
following strokes, and Reagan, due to early
symptoms of Alzheimer’s disease. Lincoln
and Truman, who received the most vicious
attacks, are now considered two of our finest
presidents. Eisenhower recovered to func¬
tion well; Wilson did not. The effect of
Reagan’s illness upon his second term is still
a matter of debate.
Concerns about intelligence are not con¬
fined to concerns about our leaders. Our
school systems use cognitive tests to stream
high school students into different pro¬
grams. Colleges use cognitive tests to screen
applicants for higher education programs.
These tests are never called “intelligence
tests,” but they correlate highly with them.
Testing is not confined to the educational
system. Volunteers for the United States
military services must obtain passing scores
on a test of general mental competence,
the Armed Forces Qualifying Test (AFQT).
Similar tests are used in many other coun¬
tries. Toward the bottom end of the scale
there are a variety of special assistance pro¬
grams for people who simply do not have the
cognitive competence to cope with the com¬
plexities of the modern world. Low intelli¬
gence test scores can be offered as evidence
of diminished mental capacity during the
penalty phase of a criminal trial.
While there is broad agreement that
some people are smarter than others, things
become more complex when we try to be
precise. I think that every knowledgeable
person would agree that Albert Einstein and
Thomas Jefferson were both highly intelli¬
gent. Who was the more intelligent? That
is hard to say; they were brilliant in differ¬
ent ways at different times. It would be easy
to find other examples of the same point.
There are clearly varieties of cognitive skill,
especially at the top. As a result some mod¬
ern observers have concluded that there is
no single dimension of intelligence.
This idea is not new. In the sixteenth
century the Spanish physician/philosopher
Juan Huarte de San Juan2 drew a remarkably
cogent picture of individual differences in
2 Huarte de San Juan, 1575/1991.
human thought. Huarte believed that when
people attack problems some will use their
imaginations to envisage how a solution
might work out, while others will rely
on their memories of solutions that have
worked in the past. Huarte also defined
“understanding” ( entendimiento ) as a sep¬
arate capacity, implying that one can be
bright without having a good understanding
of a situation. Huarte’ s distinction between
problem solving by imagination or by mem¬
ory is mirrored in contemporary theo¬
ries that distinguish between the ability to
do abstract reasoning and the ability to
apply previously learned solution methods.3
Robert Sternberg, a prolific modern writer
on intelligence, has emphasized the dis¬
tinction between analytic intelligence and
the ability to understand complex social
situations.4
Huarte anticipated another modern idea,
the need to have a biological explanation for
intelligence. Huarte offered a theory based
on the sixteenth-century notion that the
body is governed by four “humors” - blood,
bile, black bile, and phlegm. This theory of
biology has long since been discarded. The
idea that there should be a biological expla¬
nation for individual variations in cognition
has been retained. One of the most active
areas of modern intelligence research deals
with the relation between intelligent behav¬
ior and the brain.
Let us leap from the sixteenth century
to the nineteenth, and to one of the most
colorful characters in the history of sci¬
ence, the Victorian physician, mathemati¬
cian, and explorer Sir Francis Galton. Galton
explored in Africa, made major contribu¬
tions to the development of statistics, and
conducted research in psychology, most
noticeably on intelligence. He wholeheart¬
edly endorsed the theory of evolution pro¬
posed by his cousin Charles Darwin. Galton
believed that human intelligence was largely
inherited. He also maintained that intelli¬
gence was one manifestation of a person’s
overall constitutional fitness. Therefore,
5 Horn & Noll, 1994.
4 Sternberg, 2003.
THE ISSUE OF INTELLIGENCE
3
it should be possible to learn something
about a person’s intelligence by examining
his or her physique, including brain size, and
by determining the efficiency of the per¬
son’s nervous system by doing such things
as recording the speed with which he or she
reacted to a signal to strike a bag. These ideas
are alive, in much expanded form, today.
The next step was taken at the start of
the twentieth century, when the Frenchman
Alfred Binet developed the first intelligence
tests to be used in schools. Testing has dom¬
inated the study of intelligence since then,
so we need to look more closely at the idea.
1.2. Testing
If you want to go beyond saying “people
are different” you have to offer some way
of measuring those differences. There is an
imperative to develop such measures if a
society wants to assign different roles to
different people, based on their personal
characteristics. Not all societies do this, all
the time. There was no intelligence test
for the ruler in the hereditary kingships of
medieval Europe. The Hindu caste system
pre-assigned people to social roles, based
upon their birth. It is notable, though, that
both these societies experienced a good deal
of conflict due to their restricting people’s
social roles.5 6
In village-based societies personal knowl¬
edge of individuals plays a major role in
assigning people to jobs. When American
pioneers began to move into the northern
plains states Sitting Bull, the paramount
chief of the Lakota (Sioux] Indians, selected
Crazy Horse to be war leader from among
people whom he knew personally.' That
5 During the Hundred Years' War between England
and France (roughly 1350-1450) the French court was
disrupted when Charles the Foolish inherited the
throne. He probably suffered from a bipolar psy¬
chosis. In India the Sikhs were formed largely as a
protest against the rigid social structure enforced by
the Hindu caste system.
6 He did well. Crazy Horse defeated General Custer
at the Battle of the Little Big Horn. This was the
most stunning Native American victory in the his¬
tory of the western expansion.
technique does not work in today’s large
societies, where there are many positions
to be filled in both government and indus¬
tries. Leaders cannot possibly know all their
subordinates, let alone the subordinates’
subordinates. Our society requires formal
machinery for selecting candidates either
into employment, directly, or into educa¬
tional systems that serve as channels to
future employment.
Many societies solve this problem by an
elaborate form of recommendations. A boy
or (historically less often] a girl who is
thought to be talented is sent off for training
and/or apprenticeships. While the details
have been lost, this appears to be the way
the ancient Egyptians selected children to
be trained as scribes. It was also the way in
which second, third, and fourth sons were
recruited for the priesthood (or the army)
in medieval Europe. The person was not
needed at home, and somebody had connec¬
tions enough to start them on a career. The
use of connections is certainly not unknown
in modern times. But we do rely on another
method of personnel selection: testing.
There have been many objections to test¬
ing. In evaluating them it is well to keep one
thing in mind. Society needs a mechanism
for personnel selection. Not everyone can
have whatever they want. Students have to
be selected, jobs have to be filled, and when
behavioral problems arise, mental compe¬
tence must be assessed. If you do not like
testing, what is your alternative?
1.2.1. Testing Before Psychology
Modern psychologists did not invent test¬
ing. In the days when the Chinese emperor
claimed to rule “the Earth, the Moon, and
three quarters of the Sun” an elaborate
series of local, regional, and nationwide tests
was used to select officers for the imperial
bureaucracy. Candidates had to write tra¬
ditional poetry and to explain the impor¬
tance of fearing the will of heaven and know¬
ing the words of the sages. Evidently it was
assumed that a person who could do these
things could collect the imperial taxes or be
an ambassador to the Mongols.
4
HUMAN INTELLIGENCE
Some centuries later the British Empire
emulated the Chinese Empire. Until after
World War II career positions in both the
Indian empire and the British upper bureau¬
cracy were filled largely from the ranks of
people who had read history, classics, and
occasionally economics at the elite universi¬
ties, especially Oxford and Cambridge. The
British assumed that someone who could do
well on oral and written examinations of the
writings of Horace and the wars of Caesar
would have the ability to administer India or
to ferret out secrets while on Her Majesty's
Service.7
Such techniques of personnel selection
may sound quaint to us, but, on the whole,
they worked. The Chinese bureaucracy held
the empire together in a way that no suc¬
cession of able emperors could ever have
done alone. Properly educated British gen¬
tlemen administered India reasonably well
for two hundred years. The classics pro¬
gram at Cambridge University produced at
least four remarkably effective twentieth-
century spies. Unfortunately, they spied
against Britain for the Soviet Union, but
that is a motivational rather than a cogni¬
tive aberration.8 They were good at what
they did.
Why did these exercises in testing appar¬
ently irrelevant knowledge do a reasonably
good job of selecting people able to run very
large, complex empires? Or for that matter,
able to fool a modern counterintelligence
agency?
What the British and the Chinese had
stumbled on, and what we today attempt to
evaluate, was a collection of mental traits
that, collectively, we call intelligence. These
traits define individual differences in skills
that have broad application in many settings.
One of the most important aspects of intelli¬
gence is an ability to learn. You demonstrate
7 Gardner, Kornhaber, & Wake, 1996, pp. 12-16.
8 Anthony Blunt, George Burgess, David MacLean,
and Kim Philby. Burgess, MacLean, and Philby fled
to the Soviet Union when they were about to be
exposed. Blunt stayed in England, undiscovered,
and became a distinguished art historian. His espi¬
onage role, which seems to have lasted through the
1940s and 50s, was not publicly revealed until 1979.
this by showing that after exposure to
knowledge you have learned something. The
skills needed to learn the wisdom of Confu¬
cius or the philosophical ideas of Socrates
are not exactly the skills you need to run
an empire, but there is an overlap. For that
matter, the skills needed to do well on a col¬
lege entrance test are not exactly the skills
you need to acquire a bachelor's degree, but
there is an overlap. That is why both the clas¬
sic and the modern testing systems work. It
is also why they work imperfectly.
1.2.2. Alfred Binet Invents Modern
Intelligence Testing
At the start of the twentieth century the
French Ministry of Education had a prob¬
lem. The idea of universal public education
had been accepted, but the schools did not
seem to work for some students. How did
this problem arise?
France, like all modern democracies, was
(and is) committed to providing public edu¬
cation for all its citizens, so that all children
have an opportunity to compete for desir¬
able positions in society. This goal is not easy
to achieve.
Modern schooling is an historically
unusual form of education. Before 1800
most humans were educated “on the job" -
observing and then helping adults, and
serving as apprentices. Universal education,
the requirement that every child learn by
practicing seemingly esoteric exercises in
a setting divorced from everyday life, is
a late nineteenth-/early twentieth-century
idea. By 1900 it was apparent to educators
that some children have a great deal of trou¬
ble learning in this manner. The French edu¬
cational administration needed to have a
way of identifying such children, so that they
could either be dropped from the system
or channeled into an educational program
more suited to their capabilities.
The fact that France was a democracy
imposed an added constraint. French edu¬
cators needed an objective method, in pref¬
erence to the subjective impressions of
“persons in authority,” the teachers and prin¬
cipals. In an authoritarian regime there is no
THE ISSUE OF INTELLIGENCE
5
need for such a method; if the authorities
don’t like you; you’re out.
So from the very first, testing was embed¬
ded in the society that required it.9
In order to meet this challenge the Edu¬
cation Ministry hired Alfred Binet, who had
worked for a while with Galton. Binet began
by making two important assumptions. The
first was that mental competence increases
over the childhood years. The typical six-
year-old can solve problems a four-year-old
cannot; a four-year-old can solve problems a
two-year-old cannot, and so on, at least from
birth to the late teenage years. Therefore, it
makes sense to talk about mental age (MA) -
the level of mental competence at which a
child is operating.
Binet took a pragmatic approach to the
measurement of mental age. He asked expe¬
rienced teachers what sorts of problems chil¬
dren could solve at different ages. Once he
had a set of problems typical of the ones chil¬
dren could solve at age six, seven, eight, and
so on, he could determine a person's men¬
tal age by finding the most difficult prob¬
lems that a child could solve. Mental age
could then be compared to chronological
age (CA), to determine whether a child has
been performing below, at, or above the cog¬
nitive level that would be expected.
Binet then made his second, more
arguable, assumption. He assumed that a
child’s relative standing in mental devel¬
opment, compared to his age group, will
remain fairly constant as the child grows
up. If Sammy and Tommy are both six,
but Sammy has a mental age of eight and
Tommy one of five, Binet assumed that four
years later, when they are both ten, Sammy
would have a mental age higher than ten,
9 The contrast between the French and the Chinese
and British imperial systems is informative. The
Chinese and British systems were designed to select
a sufficient number of qualified candidates for gov¬
ernment functions. So long as the supply of young
officers and bureaucrats was adequate, there was
little concern that the system might have shut out
potential candidates. The French testing program
was designed to staff society, without favoring some
citizens over others. Any government has to solve its
staffing problems. Only democracies have to justify
the staffing system to the citizens.
and Tommy a mental age lower than ten.
Therefore, it follows that if you test children
on entrance to school (age six or seven), and
you find that some are markedly behind (i.e.,
have mental ages in the three-to-four range),
those children are likely to be behind their
classmates at all ages, and therefore are can¬
didates for removal from the normal school
program. That is what the French education
system wanted to know.10
The Education Ministry accepted Binet’s
argument. The modern era of intelligence
testing had been born.
1.2.3. The Intelligence Quotient (IQ)
Binet did not use the term Intelligence Quo¬
tient (IQ) because the concept of mental
age was sufficient for classifying children
who were entering school. As mental test¬
ing expanded to the evaluation of adoles¬
cents and adults, however, there was a need
for a measure of intelligence that did not
depend upon mental age. Accordingly the
intelligence quotient (IQ) was developed.
There have been changes in the definition
and use of the term since its introduction.
The details are provided in panel 1.1. Here
we proceed directly to the modern use of
the term.
The term IQ is used in two ways, which
I will call the narrow and broad uses of the
term.
The narrow definition of IQ is a score
on an intelligence test, developed accord¬
ing to a scoring protocol where “average”
intelligence, that is, the median level of per¬
formance on an intelligence test, receives a
score of 100, and other scores are assigned
so that the scores are distributed normally
about 100, with a standard deviation of 15.
Some of the implications are that:
1. Approximately two-thirds of all scores
lie between 85 and 115.
2. Five percent (1/20) of all scores are above
125, and one percent (1/100) are above
135. Similarly, five percent are below 75
and one percent below 65.
10 Binet & Simon, 1905.
6
HUMAN INTELLIGENCE
Panel 1.1. The Intelligence
Quotient (IQ)
Mental age is inadequate as a means of
comparing the intelligence of two chil¬
dren of different chronological ages (CA),
Suppose a six-year-old and a ten-year-old
both have a mental age (MA) of eight.
Cognitively, they are likely to be very
different individuals, for one is develop¬
ing rapidly and the other is developing
slowly. The need for a measure of men¬
tal development that is independent of
chronological age led to the concept of
the Intelligence Quotient, IQ, which was
originally defined as the ratio of mental
age to chronological age, multiplied by
10Q.*
MA
lQ='°°' CA-
M
To illustrate, a ten-year-old who can
solve problems at the level of difficulty
expected of a twelve-year-old would have
MA = i2, CA = 10, IQ = 120. An IQ of 100
indicates that the child’s cognitive devel¬
opment is proceeding on schedule, an IQ
above 100 indicates acceleration, and an
IQ below 100 indicates slowed develop¬
ment. In the case of our hypothetical six-
year-old and ten-year-old, both of whom
have a mental age of eight, the first child
would have an IQ of 133 and the second
an IQ of 80. In modern educational terms
the first child might be considered for
an accelerated program, while if the sec¬
ond’s IQ score were accompanied by dif¬
ficulties with schoolwork he or she would
be a candidate for a special education pro¬
gram.
This method of calculating IQ will not
work with adults, because intelligence
does not increase linearly with age past
childhood, A man of sixty whose men¬
tal powers are equal to those expected
of a forty-year-old would not be con¬
sidered a case of retarded development!
Therefore, the IQ ratio just described has
been replaced by a measure based on the
notion that IQ should reflect a person’s
relative standing within his or her own
age group.
Intelligence tests are standardized by
giving them to a relatively large sam¬
ple of people chosen to be representative
of the population for whom the test is
intended. In the case of a test intended for
broad use, such as establishing the mental
competency of adults, this is essentially
the entire population of the country, so
attempts are made to obtain a large repre¬
sentative sample of the United States, the
United Kingdom (for the British version),
Spain (for the Spanish version], and so
forth. 1 The sample should be sufficiently
large that a distribution of scores can be
obtained for different age groups.
The raw test score is based on the
number of items correctly answered
and/or the difficulty level of the item.
(The scoring algorithm varies somewhat
with the test, as discussed in Chapter 2).
The IQ score is derived from the raw
score in the following way.
Let y be a raw score for a person in
a particular age group, and let M be the
mean and S the standard deviation for all
scores in the reference group. The cor¬
responding standard score is
Y-M
The IQ score is derived from this by the
conversion
IQ = 15 • z -h 100. (1.3)
If the raw scores were normally dis¬
tributed, then the resulting IQ scores will
be normally distributed with a mean of
100 and a standard deviation of 15. The
graphic depiction of this distribution is
the famous “Bell Curve.” The Bell Curve
for IQ distributions is shown in Figure 1.1.
Why not record scores in the standard
score format? Statisticians, psychometri¬
cians, and research workers would prob¬
ably prefer to do this. The advantages
THE ISSUE OF INTELLIGENCE
7
have to be weighed against two counter¬
vailing trends: tradition and public rela¬
tions. IQ scores were introduced almost
a century ago; people are used to them.
Standard scoring is a bit esoteric for the
nonstatistician. For a statistician, having a
scale with a mean of zero and a standard
deviation of 1 is a convenience, nothing
more. If a (nonstatistical) parent were to
be told that their child had scored a zero
on an intelligence test, they might inter¬
pret this as a claim that their pride and
joy had the intelligence of a rock! Half the
people who took the test would receive
negative scores, which could lead to all
sorts of misunderstanding. Appearances
can be important.
* The concept of IQ was developed by the Ger¬
man psychologist William Stern, not by Binet.
t This procedure contains the implicit assump¬
tion that the distributions of intelligence are
the same across populations. Thus if you con¬
sider a score of 100 on the Spanish form of
the test to indicate the same thing as a score
of 100 on the American version of the test,
you are implicitly assuming that Americans and
Spaniards have equivalent intelligences, on a
population basis. Such assumptions have been
vigorously debated. The controversies are dis¬
cussed in detail in Chapter 11.
t The standard deviation is a measure of the
amount of variation in a population. More
details are given in Chapter 2.
Thus IQ, in the narrow sense, is a score
indicating a person’s relative performance
on an intelligence test, compared to the per¬
formance of people in an appropriately cho¬
sen comparison group. This does not com¬
pletely clarify the matter, because there can
be debate about what counts as an intel¬
ligence test. This matter is also discussed
on page 8. I will attempt to be clear about
how the term is being used in various con¬
texts.
In the broad sense the term IQ is used
as a synonym for intelligence, that is, as a
shorthand term for individual differences in
cognition. This can lead to confusion, so I
shall attempt to use IQ only in the sense of
a test score. The term intelligence will be used
to refer to the broader concept of individual
differences in mental ability. In my usage, a
person who has high intelligence will proba¬
bly have a high IQ score, but the distinction
between the two is important.
In interpreting IQ scores it is often use¬
ful to think of percentiles, which indicate
the percentage of people in the referent
group whose scores are below a certain level.
What that level is will be determined by the
IQ score and by the properties of the Bell
(normal) Curve itself. Table 1.1 gives some
important reference scores. The properties
of these scores follow from the assumption
that IQ scores will follow the normal distri¬
bution, which is illustrated in Figure 1.1.
As a result, in terms of the modern scor¬
ing, if someone says that their child has an
IQ of, say, 120, this does not mean that the
child’s mental age is 20% higher than his
chronological age. It means that the child
has a test score in the top 9% of test scores
at the child’s age.
Why can we assume that IQ scores are
distributed normally? The answer is simple.
IQ tests (and many other tests) are con¬
structed by choosing appropriate numbers
of easy, intermediate, and hard items, so
that the total scores will be normally dis¬
tributed in the population for which the test
was intended. The Bell Curve is an artifact
of the way the test is constructed! There is
no definition of IQ independent of the tests
themselves. This contrasts with a variable
like height, which is defined independently
of yard sticks or meter sticks. Height hap¬
pens to be distributed approximately nor¬
mally, within the populations of adult males
and females. The distribution of height is a
fact of nature. The fact that IQ test scores
are normally distributed is an outcome of
the test construction procedure. Neverthe¬
less, it is a reasonable thing to do. Why?
IQ scores are used to describe people rel¬
ative to each other. They are also used to
8
HUMAN INTELLIGENCE
Table 1.1. The distributions of standard and IQ scores in
terms of the percentage of people above or below selected
scores. An IQ of 65 would, if accompanied by other
indications of mental incompetence, be cause for considering
a person mentally disabled. If IQ is distributed normally,
about one percent of all people have IQ scores this low.
Average IQ is, by definition, 100. Approximately half of all
scores lie between 90 and 110. About 16% of the scores lie
above 115, 2% above 130, and 1% above 135. MENSA , an
organization whose members have high IQ scores, defines
the 4 sigma group, people with IQs over 160. ( Sigma is a
term frequently used to refer to the standard deviation.) This
level of score would be expected three times in every 100,000
observations.
Standard Score (z)
IQ Score
% Below
% Above
-2-33
65
0.982
99.018
—2.00
70
2.275
97-725
—1.00
85
15.866
84.134
-.67
90
25.249
74-751
0.00
100
50.000
50.000
.67
110
74-751
25.249
1.00
u5
s4-134
15.866
2.00
130
97-725
2.275
2-33
135
99.0l8
O.982
3.00
M5
99.865
O.I35
4.00
160
99.997
O.OO3
make predictions and to indicate associa¬
tions, as in predicting a student's likely aca¬
demic progress or investigating the associa¬
tion between intelligence and income. There
are technical reasons for wanting to deal
with normally distributed scores when we
apply the statistical methods used for mak¬
ing predictions and analyzing associations.
There is another, less technical reason for
requiring that IQ scores be normally dis¬
tributed. Many other human qualities that
can be measured on scales with physical
interpretations, like height and weight, are
distributed normally. It seemed to many of
the early researchers that if we could mea¬
sure intelligence in some physical manner,
such as measuring the efficiency of the ner¬
vous system, these measures would probably
turn out to be normally distributed. There¬
fore, it seemed appropriate to require that
IQ scores be normally distributed.
In the late nineteenth and early twentieth
century this reasoning seemed compelling,
because the normal distribution itself was
regarded (almost mystically) as a Law of
Nature. Today we are a bit more skepti¬
cal, but there is still a good argument for
assuming normality. If a person’s intelli¬
gence is due to a large number of indepen¬
dent causes, each of which has a small effect,
intelligence would be distributed normally
across the population.
A certain amount of the confusion
between the broad and narrow senses of IQ
is due to the way in which cognitive tests
are described. Some tests are explicitly mar¬
keted as intelligence tests. But because the
term IQ, and sometimes even the term intel¬
ligence, have acquired a bad taste in cer¬
tain circles, many tests of cognitive skills
are not marketed as intelligence tests, even
though these tests are highly correlated
with tests that are marketed as intelligence
testsl For instance, in a widely read and
highly controversial report, Richard Herrn-
stein and Charles Murray used the Armed
THE ISSUE OF INTELLIGENCE
9
Figure 1.1. The "Bell Curve" for IQ. The area under this curve
represents 100% of the population. The area under the curve and to
the left of a given IQ value (on the abcissa) represents the fraction
of people in a population who have IQs lower than the indicated
IQ. Conversely, the area to the right indicates the fraction of
people who have this IQ or a higher one. For example, 50% of the
area under the curve lies to the left of IQ = 100, indicating that half
the population has an IQ of less than 100. Nine percent of the area
under the curve lies to the right of 120, indicating that only nine
percent of all people have IQs of 120 or higher. The Bell Curve for
IQ scores is a special example of the normal, or Gaussian,
distribution. At the extremes, the curve never quite touches the
abscissa (“x axis”), but this cannot be shown on the graph.
Services Qualifying Test (AFQT) as a mea¬
sure of intelligence, and treated AFQT and
IQ scores as being virtually synonymous.11
The US Department of Defense never refers
to the AFQT as an intelligence test. Similar
confusions arise with the SAT. Many
research projects have used SAT scores as a
measure of intelligence, although the test’s
publisher, the Educational Testing Service,
does not describe it as an intelligence test.
There is great controversy over whether
or not IQ scores should be treated as
real indicators of mental ability. Panel 1.2
presents a historical debate that took place
in the 1920s, but in many ways foreshadowed
contemporary arguments. I shall come down
squarely in the middle of the controversy. I
will argue that the scores certainly do mean
11 Hermstein & Murray, 1994.
something, but they may not mean as much
as some enthusiasts claim.
1.2.4. WTnzt Binet Discovered: “ Drop
in from the Sky " Testing Works
Let us take a closer look at what Binet
assumed and what he found.
Binet’ s assumption that mental compe¬
tence increases as children grow older is
certainly correct. Mental competence may
decrease in old age, but that is another story,
and was of no concern to Binet. He was also
correct that there are marked individual dif¬
ferences in the rate at which mental compe¬
tence increases.
His second assumption, that relative
standings remain constant as children age, is
true on the whole, but there are exceptions.
As a toddler, Albert Einstein was a relatively
10
HUMAN INTELLIGENCE
Panel 1.2. Defining Intelligence: The
Debate between Mr. Lippmann and
Professor Boring
In science clarity of definition is essen¬
tial, for good definitions make clear what
the important questions are. The study of
intelligence has been plagued by a lack of
precise definitions. The debate between
Lippmann and Boring, early in the twen¬
tieth century, shows how the failure to
define terms introduced confusions that
continue to this day.
Following the use of tests in World
War I, intelligence testing became
a growth industry. So, inevitably, it
attracted the attention of learned com¬
mentators - people who, if there had
been TV in those days, would have
appeared as talking heads on the Sunday
morning pundit shows. One of the most
respected of these commentators, Wal¬
ter Lippmann, did not at all like the new
technology. He was particularly incensed
by a claim, based on analyses of the Army
Alpha data, that the average American
had a mental age of fourteen. In Lipp-
mann's own words:
The intelligence test, then, is an instru¬
ment for classifying a group of peo
pie, rather than 11 a measure of intel¬
ligence .n People are classified within a
group according to their success in solv¬
ing problems which may or may not be
tests of intelligence . They are classified
according to the performance of some
Californians in the years 1910 to about
iqi6 with Mr. Terman’s notion of the
problems that reveal intelligence. They
are not classified according to their abil¬
ity in dealing with the problems of real
life that call for intelligence .
(Lippmann, 1922a)
Lippmann argued that the test devel¬
opers had produced a barrage of statis¬
tics that had the trappings of science,
but were not scientific. The tests them¬
selves were based on hunches by people
such as Terman that this or that behavior
indicated intelligence, rather than on
any scientific theory of what constituted
intelligence. Lippmann also doubted that
a classification of people on the basis of
test scores would map onto a classifica¬
tion according to their ability to “deal
with problems of real life that call for
intelligence.”
Academia responded. The Harvard
professor E. G. Boring* clarified the mat¬
ter by asserting:
Intelligence is what the intelligence tests
test.
(Boring, 392^
So therel
The exchange between Lippmann and
Boring foreshadowed a debate that is
active today. Should a test be developed
inductively, by the pragmatic procedure
of identifying people who are believed to
vary in intelligence, seeing what behav¬
iors distinguish those who are intelligent
from those who are not, and then incor¬
porating these behaviors into a test? Or
should a test be based on a theory of how
individual differences in cognitive power
come to arise?
The answer is not a simple one. To
see why, consider the following analogy.
I think that everyone would agree that
people differ in their physical fitness. But
what is physical fitness? You could take
the approach of an athletic coach, and ask
people to run, jump, throw weights, and
so forth in order to determine physical fit¬
ness. Alternatively, you could take a med¬
ical approach. Physical fitness depends
on muscular adequacy, reaction, and the
ability of the heart and lungs to provide
fuel to the muscles. So let us take mea¬
sures of cardio-pulmonary capacity and
construct tests of the strength of isolated
muscle groups and the speed of neural
impulses.
Binet, Terman, and their many suc¬
cessors took the coach’s approach. Lipp¬
mann seems to have wanted a more the¬
oretically justified approach, although he
THE ISSUE OF INTELLIGENCE
ll
did not offer one. (After all, he was a jour¬
nalist, not a scientist.) Lippmann did not
deny that something called intelligence
exists, for he spoke of dealing with prob¬
lems that call for intelligence. His objec¬
tion was to the tests offered to evaluate it.
Boring’s response, identifying intelli¬
gence with a test score, strikes many
(including me) as somewhat arrogant, for
it confounds the concept of intelligence
with a score on an imperfect indicator of
intelligence. However, it does lead to a
research agenda.
The first item on the agenda is to ask
whether or not the tests identify peo¬
ple who would otherwise be considered
intelligent. Lippmann said they would
not, but that was his opinion, not an
observation of fact.
To some extent the way in which
the tests were developed ensured that
they did identify “the intelligent.” Binet
and Terman began with groups who had
known variations in cognitive compe¬
tence (e.g., younger and older schoolchil¬
dren), and then developed tests whose
scores reflected the variation. Lippmann
pointed out that this procedure makes
the definition of intelligence dependent
upon an appropriate choice of exami¬
nees for the initial (. standardization ) test¬
ing. This was an important observation.
Today there is a lively debate over the
appropriateness of using intelligence tests
that have been standardized in the post¬
industrial world to evaluate intelligence
in the developing nations of Africa, Asia,
and Latin America.
But Boring had a point, too. If test
scores are accepted as good indicators of
intelligence, the tests become a power¬
ful tool for investigating individual dif¬
ferences in cognition. You can determine
how test scores are related to variations
in brain structure and/or process, genetic
makeup, schooling, family support, and a
variety of other causal factors. Test scores
can also be used as predictors of success
or failure in all walks of life. We can
then consider different explanations for
observed relationships. Science advances
by accounting for relationships between
observables, not by debating beliefs about
what ought to be the case.
Boring's view has lead to the psycho¬
metric approach to intelligence. It has
resulted in a viable research agenda. But
it contains a weakness.
If intelligence is equated with an IQ
score, the study of intelligence is reduced
to an analysis of the causes and conse¬
quences of test scores. So if the origi¬
nal process of test construction missed
something important about intelligence,
the assertion that intelligence is what the
tests test makes it hard to incorporate
new perspectives. Lippmann was right to
be concerned over the conceptual limita¬
tions of a fascination with test scores.
The argument between Lippmann and
Boring foreshadowed future develop¬
ments. There are many advocates of
the psychometric approach today. It is
not entirely devoid of theory. In fact,
the analysis of pragmatically defined test
scores has suggested theoretical posi¬
tions about intelligence that can then
be expanded to include new measures.
Some of these efforts will be presented in
later chapters.
There are also contemporary psychol¬
ogists who accept Lippmann ’s concerns
(although he is seldom cited!) and then
go beyond criticizing to try to develop a
theoretical basis for intelligence and intel¬
ligence testing. Such a theory will be use¬
ful if it leads to a coherent picture of
how individual differences in cognitive
power arise and influence human expe¬
rience from birth to death.
In Section 1.4 I present my own views
about what such a theory has to deal with,
and explain the difficulties that it encoun¬
ters.
* Boring, 1923.
12
HUMAN INTELLIGENCE
late talker! Most of us can think of some¬
one who was not terribly impressive in grade
school or high school, but who had a great
college career. Things go the other way, too.
The smartest kid in grade school does not
always become a Phi Beta Kappa in college,
let alone becoming a prize-winning physicist
or a fabulously wealthy entrepreneur.
Such examples are striking, but they are
exceptions to a general rule. Past the age
of about ten, indicators of relative cogni¬
tive competence are fairly stable. A Scot¬
tish study found a substantial correlation
between intelligence test scores taken at
age eleven and subsequent measures taken
when the examinees were in their sixties
and seventies.12 In psychological jargon intel¬
ligence is a trait, a characteristic of the indi¬
vidual that is reasonably stable over time and
that is revealed in many situations.
Binet’s third assumption was the unno¬
ticed 6oo-pound gorilla in the room. Let us
accept that there are individual differences
in the mental competencies required to be
successful in modern society.15 These com¬
petencies, collectively, are what we mean
when we say “intelligence.” Binet assumed,
and then showed, that it is possible to make
reliable measurements of a substantial num¬
ber of the traits that constitute intelligence,
within the space of three or four hours.
I suggest that the reader go back and
weigh each clause of that sentence. Binet
showed that some of the cognitive traits
important to success can be evaluated within
a single testing session, conducted outside
of the context of everyday activities. I like a
term originally coined by Robert Mislevy,^
Drop in from the Sky testing, because it cap¬
tures the examinee’s view of the out-of-
context nature of the assessment. Much of
12 Deary et al., 2000.
13 Individual differences in mental competence prob¬
ably occur in every human society, although the
relative importance of specific competencies may
vary from society to society. The ability to locate
one’s self in space is less important in a society that
has street signs and global positioning systems than
in a society of hunter-gatherers.
14 Mislevy, formerly at ETS and now a professor at
the University of Maryland, is a highly respected
specialist in the field of psychometrics, the mathe¬
matical analysis of test scores.
the research in the century following Binet
can be looked upon as an attempt to refine
and expand the measurements that can fit
into the “Drop in from the Sky” paradigm.
Throughout this book I shall argue that
by accepting the conventional testing para¬
digm as a given, intelligence researchers have
too often ignored traits that cannot be
measured within the paradigm but that,
by any reasonable definition, are part of
intelligence. Nonetheless, a nontrivial sub¬
set of human cognition does fit into the
“Drop in from the Sky” paradigm, and hence
can be evaluated using conventional testing
methods.
1.2.5. Expansions ofBinet's Work : The
Stanford-Binet and Wechsler Tests
Binet’s tests excited a great deal of inter¬
est. Educationally, the need for classifying
students was a problem for the burgeon¬
ing public educational systems of the early
twentieth century. Intellectually, the tests
provided a foothold for applying the new,
but rapidly developing, science of psychol¬
ogy. Lewis Terman, a professor at Stanford
University, translated and modified Binet’s
tests for use in the United States. Due to
the interruptions caused by World War I,
Terman did not complete his work until the
1920s. The resulting test, the Stanford-Binet
Intelligence Test, is still used in regularly
updated form today.
The Binet and Stanford-Binet tests
were intended for use with schoolchildren,
roughly age fifteen and younger. In the late
1930s David Wechsler, a clinical psychologist
working at New York City’s Bellevue Hos¬
pital, created a similar test for adults, the
Wechsler-Belleime test. It has subsequently
been modified into the Wechsler Adult Intel¬
ligence Scale (WAIS). It and a companion
test for children, the Wechlser Intelligence
Scale for Children (WISC), are the most
widely used individually administered intel¬
ligence tests today. The Wechsler tests are
often referred to as the “gold standard” for
intelligence tests.15 As of 2010 the fourth
15 Matarazzo, 1972.
THE ISSUE OF INTELLIGENCE
*3
revisions of the tests, WAIS-IV and WISC-
IV, were in use.
Both the Wechsler and the Stanford-
Binet tests are individually administered.
The examinee sits down with a trained
examiner and attempts to solve a series of
problems, divided pragmatically into prob¬
lems that do or do not involve language, and
that vary in the demands that they place on
memory. Wechsler has described this as an
opportunity for the examinee to display his
or her cognitive skills during a standardized
interview with an experienced observer.16
The resulting IQ scores have proven to be
highly useful, both in the educational system
and in other settings. For instance, the WAIS
is widely used to evaluate a person who, for
whatever reason, is of suspected mental dis¬
ability. Examples of such use are the adjudi¬
cation of legal competence and the analysis
of status following brain injury. Other appli¬
cations of these tests, and there are many,
are extensions of these ideas. The focus is
always on the individual. The cost of testing
is evaluated relative to the potential benefits
of the results in making judgments about
an individual case. Are the decisions made
about this person improved by knowing test
scores, and if they are, is the value of a typ¬
ical decision enough to justify the costs of
the test?17
The Wechsler and Stanford-Binet tests
are not the only individually adminis¬
tered intelligence tests. However, they have
played a very important role in the develop¬
ment of testing. Some further tests will be
discussed in Chapter 2, where the tests are
described in more detail.
16 Wechsler, 1975.
17 The law provides a dramatic example. The United
States Constitution prohibits “cruel and unusual
punishment” but does not define it. Several of the
states and the current federal code include capital
punishment for treason and for particularly vicious
cases of murder. The Supreme Court has held that it
would be cruel and unusual punishment to execute
a mentally retarded person for committing a capi¬
tal crime, on the grounds that the person, though
guilty, could not understand the crime or its con¬
sequences. As of 2009 a defendant’s attorney could
offer an IQ test score as one piece of evidence of
mental retardation and, hence, as a reason for not
assessing the death penalty.
1.2.6. The Development of Group Testing
The next major step in intelligence testing
was a spin-off of technology from a military
application. When America entered World
War I the army had to make rapid men¬
tal evaluations of large numbers of incom¬
ing soldiers. The War Department18 spon¬
sored development of a test that could be
administered to large groups of recruits.
Psychologists responded with the Army
Alpha Test, a written test suitable for group
administration, and the Beta test, which
could be given to nonliterates.19
The program was considered a success.
Today the militaries in virtually all devel¬
oped nations routinely use cognitive tests to
screen recruits. The tests that the U.S. mili¬
tary uses will be described in Chapter 2.
The military tests are examples of per¬
sonnel classification tests. Cognitive tests for
personnel classification are widely used in
the civilian sector as well as in the military.
The costs and benefits of testing within a
personnel classification system are not the
same as the costs and benefits of testing
intended for individual counseling and/or
placement. In a personnel classification sys¬
tem correct classifications have a value and
incorrect classifications have a cost, as seen
from the perspective of the institution set¬
ting the test, rather than as seen from the
perspective of the examinee. A classifica¬
tion test is economical if, on the average,
the cost of administering the test is less than
the value of improved decision making. This
view shifts the focus from decisions about an
individual to the average value of a decision,
calculated over the entire population. The
shift greatly affects the economics of testing.
A cheap test, which makes only a moderate
improvement in the accuracy of the selec¬
tion decisions, administered to thousands of
or even millions of people, can be a valuable
classification instrument.
18 Now renamed the Department of the Army and
absorbed into the larger Department of Defense.
19 In the United States today illiteracy is almost a signal
that the individual is of suspect intelligence. That
was not true in 1917-18, for universal public edu¬
cation was not the case. Many people of normal
intelligence were illiterate, especially in rural areas.
H
HUMAN INTELLIGENCE
Well over a hundred group-administered
classification tests have been developed.
They include the Scholastic Ability Test
(now renamed by its acronym, the SAT]
used in the college admissions process, and
the General Aptitude Test Battery (GATB),
which for years was used by the United
States Department of Labor to provide a test
score to guide in industrial hiring, and many
more.
Describing all the tests in use today liter¬
ally takes a volume, and the volume has to be
updated annually. Some of the more promi¬
nent tests are described in detail in Chap¬
ter 2. The important point is that since the
1930s intelligence testing has been widely
used to make important decisions about
people’s academic and vocational careers.
Testing is also used as a guide in medical
rehabilitation, such as evaluating the course
of treatment following insults to the brain.
The tests are also widely used in research on
the description, causes, and consequences of
being intelligent.
1.3. Do the Tests Work? Efficacy
and Controversy
There has been a great deal of debate about
how well intelligence tests work. The debate
is not surprising, because the issue is inher¬
ently complicated. How accurate does a test
have to be before we say that “it works’?
There can be different standards for differ¬
ent purposes. How well do we understand
why the tests work? People are uncomfort¬
able using indicators that they do not under¬
stand. To the scientist interested in intel¬
ligence, though, this question ought to be
a challenge. What are the consequences of
basing important decisions about education,
employment, and personal planning on test
scores?
We will look at the question of test accu¬
racy in considerable detail in Chapter 10.
Here I will do three things: explain the cri¬
terion by which tests are judged as being
accurate or not, provide just a few interest¬
ing statistics (saving the rest for Chapter 10],
and then describe the controversies.
Tests would be perfectly accurate if,
whenever a test score was used as a predic¬
tor, there was some critical score such that
everyone who had a score equal to or higher
than the critical score succeeded, and every¬
one with a lower score failed. No such intel¬
ligence test exists, and none ever will. Test
scores are not perfect indicators of a person’s
cognitive power; other things besides intel¬
lectual talent determine success; and success
itself is not an either/or thing. The question
is not whether or not test scores can be used
to make perfect predictions of success or
failure; the question is whether or not using
test scores reliably improves our ability to
predict who will succeed or fail. Tests that
do this are said to be valid . Validity is a mat¬
ter of degree; the more using a test improves
prediction, the higher the test’s validity.
Here is an example that illustrates the
issue.
In the 1930s the Scottish psychologist
Godfrey Thomson conceived the impressive
plan of testing virtually the entire nation of
Scotland. And he managed to do it. In 1932
Thomson’s Moray House intelligence test
was given to almost every eleven-year-old
in Scotland, more than 80,000 in all.
About seventy years later Ian Deary, a
professor at the University of Edinburgh,
and his colleague Lawrence Whalley traced
2,300 of the people whom Thomson had
tested. Figure 1.2, taken from their report,
shows the fraction of the original respon¬
dents to the 1932 test who were still alive
at different times from the 1930s until the
start of the twenty-first century. The data is
shown separately for men and women from
the upper quartile (top 25%) and lower quar-
tile (bottom 25%] of the distribution of test
scores. Intelligent people live longer!
An intelligence test score obtained in
childhood was a valid statistical predictor
of length of life. This ought to catch the
reader’s attention.
Now let’s use this fact to uncover some of
the complications dealing with test scores,
and by implication, intelligence itself, as a
predictor.
We first must determine whether or not
we are dealing with an artifact due to
THE ISSUE OF INTELLIGENCE
*5
I
"to
Age (years)
100
90
80
70
60
50
40
30 ,
□
□
□
Men - highest IQ quarter ^
10 20 30 40 50 60 70 80
Age (years)
Figure 1.2. The fraction of individuals surviving to various ages for
a cohort of approximately 2,200 Scottish children tested at age
eleven, in 1932. From Whalley and Deary, 2001, Figure 1. By
permission of the authors and the British Medical Journal
some relationship that is forced between
test scores and the outcome of interest. For
instance, in the United States the SAT scores
of people who enter college are higher than
the test scores of people who do not enter
college. That fact, alone, is uninteresting,
because test scores are one of the determin¬
ers used to decide who gets to go to col¬
lege. However, this sort of problem did not
arise in the Scottish study, for it is unlikely
that a score on a test taken at age eleven
has any direct influence on mortality. The
statistic indicates some meaningful relation¬
ship between mortality and whatever the
test measures. But what?
I can imagine three different reasons why
a test score might predict mortality. There
might be a direct relationship. We know,
today, that intelligence test scores are partly
determined by the state of the brain. This
statement will be documented in detail in
Chapters 6 and 7. It could be that the test
scores (imperfectly) revealed the state of the
brain at age eleven, and this state carried
forward over the years, producing an associ¬
ation between test scores and later mortal¬
ity. This explanation would have appealed
to both Galton and Juan Huarte de San Juan.
There could be an indirect relationship.
Perhaps intelligence, as revealed by the tests,
makes people less likely to do unhealth¬
ful things, like drinking alcohol to excess,
or driving a motor vehicle after drinking
to excess, or forgetting to take vaccinations
against influenza, and so on. This explana¬
tion - that intelligence is related to finding
smart solutions as you navigate your way
through life - would have appealed to Binet.
Finally, it might be that the relationship
is not due to intelligence at all; it is due to
a third variable that both causes test scores
(although it is not intelligence as we nor¬
mally think of the term) and also influ¬
ences mortality. Parental socioeconomic sta¬
tus (Parental SES) is a term used to refer
to the general “place in society” occupied
i6
HUMAN INTELLIGENCE
by a child’s family. It includes such things
as financial status, education, and generally
beneficial lifestyle practices, such as hav¬
ing children inoculated against common dis¬
eases. It is also an indicator of the extent to
which a child is likely to have “family con¬
nections” to help establish his or her own
lifestyle. There is a positive statistical rela¬
tionship between children’s test scores and
Parental SES. It might be that childhood
test scores predict mortality because
Parental SES assists in the development of
a lifestyle that promotes longevity (or the
opposite, for low test scores and Parental
SES). The contemporary American psychol¬
ogist Richard Nisbett, a professor at the
University of Michigan, has made a strong
argument that many of the relationships
between test scores and social outcomes are
actually relationships between SES and the
outcomes.20
Which of these explanations do you think
accounts for the Whalley and Deary data?
Could two or more explanations both be
true?
There are probabilistic relationships bet¬
ween test scores and a variety of interesting
outcomes in life. In general, the higher a per¬
son’s test score, the more likely that person is
to have good things happen to them. These
include educational achievement, on-the-
job performance, income, marital status,
health and longevity.21 Most of the statisti¬
cal relationships are high enough to be eco¬
nomically valuable evidence to guide deci¬
sion making, both in personal life planning
and in personnel classification. But because
the relationships are probabilistic they have
to be understood in statistical terms. This
demands some expertise on the part of the
person who tries to interpret the facts.
Test scores are not only correlated with
outcomes in life, they are also correlated
20 Nisbett, 2009.
21 There are a great many studies backing up these
assertions. See Neisser et al. (1996) for the state¬
ment of the situation by a review panel of the
American Psychological Association, Gottfredson
(1997, 2007b) for reviews, and Herrnstein and Mur¬
ray (1994) for a comprehensive analysis involving
the AFQT. Chapter 10 discusses the topic in much
more detail.
with a variety of other measures that are
correlated with the outcomes. The Parental
SES example just given is illustrative. It
is often hard to determine why there is a
relationship between intelligence test scores
and outcomes such as educational achieve¬
ment, job performance, and health. The
fact that there is a correlation between an
IQ score and X, where X is almost any
life outcome, does not prove, alone, that
“intelligence causes X.” There will almost
always be several plausible interpretations
for the relationship, and they are often not
mutually exclusive. In the intelligence-life
expectancy example three alternative expla¬
nations were offered, and none of them
excluded the others. The questions “What
causes intelligence?” and “What does intelli¬
gence cause?” are not easily answered.
Do not expect simple answers about
intelligence in this book. I shall try to be as
clear as I can, but some things are necessarily
complex.
If you encounter simple answers some¬
where else, my advice is to be very, very
suspicious.
1.3.1. The Controversies
Because test scores are used to make major
social decisions, it is hardly surprising that
they have been controversial. One of the
first of these debates occurred in the 1920s,
when the social commentator Walter Lipp-
mann published a series of articles criticiz¬
ing intelligence testing in The New Republic,
an influential magazine in liberal intellectual
circles. Lippmann’s articles provoked a furi¬
ous response by Terman and, in somewhat
more measured terms, a critique by the Har¬
vard professor E. G. Boring. Boring’s reply
included the somewhat famous definition of
intelligence as whatever the tests test.
The debate between Lippmann and Bor¬
ing is described in Panel 1.2. Their argument
touched on concerns about intelligence test¬
ing that are still raised today. Other con¬
cerns have also been raised. Most of these
concerns deal with complicated facts about
what intelligence is worth, and who has
it - topics that are discussed in detail in
THE ISSUE OF INTELLIGENCE
*7
Chapters 10 and 11. Here I will foreshadow
the discussion of the controversies briefly,
by introducing the objections to the tests
and outlining the nature of the responses.
Objection 1. The tests cannot possibly
work. It is unreasonable to believe that
performance on a " Drop in from the Sky”
test, made up by people who do not
know the examinee , could possibly reveal
important mental traits.
A slightly more sophisticated way of say¬
ing this is that the objector believes that
human thought is so subtle that its nature
could not possibly be captured in a single
interview. This makes the objection seem
much less anti-intellectual, but it amounts
to the same thing as my original, italicized
version.
I do not know whether to call Objection
1 a Know-Nothing attitude (I don't want
to learn) or a Know-All attitude [I already
know the answer, so your tests are not neces¬
sary). In any case, this objection dominated
the debate between Lippmann and Boring
(Panel 1.2), and it still arises today. I do not
believe that it is productive.
Objection 2. The tests don't work. There
are people who have only modest test
scores and do well , and people who have
very high test scores and are not doing
notably well.
Here the objector asks that all predictions
made on the basis of test scores be accurate.
This is an impossible goal. Intelligence is not
the only thing that determines success or
failure, in academics or in life in general.
I conjecture that the objection is rooted
in two facts - a psychological one and a soci¬
ological one.
The psychological fact is that when peo¬
ple have to deal with probabilistic relation¬
ships they try to get an intuitive feeling for
the relationship, rather than analyzing num¬
bers. This leads to an overinterpretation of
exceptional cases. Because the tests do work
pretty well, on the average, failures stand
out. People remember the case of the person
with low test scores who did brilliantly, or
the one with high test scores who did mis¬
erably. The fact that most people perform
about as well as their test scores predict,
with a small amount of variation, just does
not stick in memory.
The sociological fact is that many of us
live in a cognitively stratified society. Chil¬
dren of parents with high, middle, or low
SES go to schools with children of the same
background as their own. Academic test¬
ing mechanisms stratify college students on
the basis of SAT or similar test scores. In
the workplace our coworkers tend to be
of roughly the same intelligence as we are.
When we deal with people of greatly varied
intelligence we tend to do so in a stereotyped
way. A sales clerk, for instance, encounters
customers of highly varied intelligence, but
does not interact with them in a way that
would reveal their intelligence. Stratification
occurs at home, because there is a great deal
of residential segregation by socioeconomic
status; therefore, neighborhood life is “strat¬
ified by intelligence."
The result is that within a person’s local
society there will not be a great deal of varia¬
tion in intelligence, so other factors, such as
variations in personality, will play a role in
determining social success. Therefore, when
people try to evaluate intelligence by refer¬
ring to their personal experiences they are
likely to undervalue the role of intelligence
in the general society.
The poor statistician has an uphill climb
when trying to show that intelligence really
is an important part of the big picture.
Objection 3. The tests work only in the
academic arena.
This objection flies in the face of the evi¬
dence. The point will be addressed directly
in Chapter 10, where evidence will be given
showing that test scores predict success
in both the academic and the industrial/
economic sphere. Because the argument has
to be made statistically, it has to counter the
results of the intellectual stratification of the
workplace, as described in the response to
objection 2.
i8
HUMAN INTELLIGENCE
There is also another problem. Leav¬
ing aside spectacular cases, like billionaires
heading companies that pioneer new tech¬
nology, it is hard to establish who is suc¬
ceeding in the workplace on an individual
basis. This makes the statistical analyses
more complex than they are in the case
of analyzing academic success, and hence
harder to explain to the nonstatistician.
Objection 4. The tests work only for
certain demographic groups, notably
middle-class whites. The tests do not pre¬
dict well for other groups.
This issue will be addressed directly in
Chapter 11. Foreshadowing, while some data
is lacking, it is usually found that test
scores have about the same power to predict
achievement in all demographic groups.
Objection 5. The tests should not be
used because they are prejudiced against
minority group members, who tend to get
low scores.
This objection raises one of the most
incendiary topics in psychology: the possi¬
bility that there are individual differences in
intelligence across racial and ethnic groups.
The topic is discussed in detail in Chapter 11.
Here I present a brief statement of the issue.
In the United States, and in other indus¬
trially developed nations, there are differ¬
ences in the average intelligence test scores
of different racial/ethnic groups. In the
United States the order of scores, from
highest to lowest, is Asian-derived groups,
European-derived groups, Latin-American-
derived groups, and then African-derived
groups. The Asian-European gap is reliable,
but rather small. The European-Latino and
European-African gaps are substantial. The
debate is not over the existence of differ¬
ences, it is over their meaning and their
implications for action. These are two sepa¬
rate issues.
If the tests predicted performance only
in the white group, discrepancies in test
scores across groups could simply be disre¬
garded. However, that is not the case; the
test scores do predict performance in all
racial/ethnic groups. Although many strong
opinions have been expressed, there is no
consensus about the causes of group dif¬
ferences in IQ scores. Because this debate
requires a good understanding of intelli¬
gence itself, I have postponed the discus¬
sion of racial and ethnic differences until the
next-to-last chapter.
Should test scores be used in person¬
nel screening if doing so would reduce
the proportion of minority applicants who
are accepted for employment or education?
There is no justification for using a test that
would have an adverse impact on one group
or another if that test is not a valid predictor
of performance. If the test is a valid indi¬
cator - as intelligence tests usually are, to
some degree - then the policy maker is faced
with a trade-off. Should the best people be
selected, regardless of group membership,
or should some effort be made to balance
rates of acceptance across racial and ethnic
groups? This is a policy issue, not a scientific
one. Scientific research can provide informa¬
tion about the costs and benefits of a policy,
but the decision is up to the policy maker.
1.4. A Framework for Thinking
about Intelligence
1.4.1. Manifest and Latent Variables:
Definition and Diagramming Conventions
Many of the arguments over intelligence
have been less informative than they might
be, because the arguers have not been clear
about what their concept of intelligence is,
or about what they think the relation is
between test scores and intelligence, in the
more general sense of individual differences
in mental competence. Here I present my
own views. They represent an expansion of
a model of the causes and consequences
of mental competence that I developed
together with Jerry Carlson, a professor at
the University of California, Riverside.22 We
did not present a new theory of what intelli¬
gence is, nor do I here. The goal is to provide
22 Hunt & Carlson, 2007a.
THE ISSUE OF INTELLIGENCE
*9
Time
Figure 1.3. A model of the development and
utilization of cognitive abilities (“Intelligence”).
Intelligence is a latent variable, indicated by
manifest variables that include but are not
limited to intelligence test scores. Intelligence
influences, and is influenced by, a number of
environmental and biological variables.
a framework for discussing theories and facts
about intelligence.
Figure 1.3 presents the model I will use
throughout this book. Before discussing its
content a word is in order about the notation
used in the figure, for these notational con¬
ventions will be followed throughout this
book.
Rectangles will be used to represent man¬
ifest variables, things that can be seen and,
in principle, measured. Test scores, grades,
and salaries are examples of manifest vari¬
ables. Ellipses will represent latent variables,
conceptual entities that are used in theories.
Intelligence, cultural knowledge, and socio¬
economic status (SES) are latent variables.
They can be defined only by their (imper¬
fect) manifest indicators. Latent variables
are never defined or measured directly; their
values are inferred from observation of the
appropriate manifest variables. For example,
SES is inferred from education, wealth, and
residence. Intelligence is inferred from test
scores and, in many cases, other indices of
cognition.
We also have to distinguish among three
separate relations between variables: causa¬
tion, reciprocal causation, and correlation.
Causation will be indicated by a single¬
headed arrow. For instance, in Figure 1.3
there is a single-headed arrow between
genes and brain structure, because genetic
makeup does determine brain structure.
Reciprocal causation, in which one variable
causes another and then the second feeds
back into the first, will be indicated by a
double-headed arrow. The double-headed
arrow between intelligence and the social
environment indicates that a person's intel¬
ligence partly determines his or her social
environment, which in turn influences the
further development of intelligence. Corre¬
lation, in which two variables tend to occur
together, without any implication of causa¬
tion, will be indicated by a double-headed
arrow with a dashed line between the arrow¬
heads. There is an example of this in Fig¬
ure 1.3, and we will encounter examples
of correlation without causation in other
diagrams.
The debate between Lippmann and
Boring can be cast in terms of manifest and
latent variables. Lippmann thought of intel¬
ligence as a latent variable, but failed to grap¬
ple with the question of how it should be
related to manifest variables. Boring either
did not make the distinction, or wanted to
assert that his manifest variable, the test
score, was a perfect indicator of the latent
variable. I do not think that either position
can be maintained.
1.4.2. The Causes of Intelligence
Now let us turn to substance, working from
"the beginning,” at the left side of Figure 1.3,
toward the expression of intelligence repre¬
sented on the right-hand side.
It all starts with the genes. A person’s
genetic makeup determines his or her poten¬
tial for the development of brain structures
that support all activities, including cog¬
nition. There are individual differences in
genetic makeup; otherwise we would all be
clones. These differences do have implica¬
tions for the development of intelligence,
even though most people probably operate
well below their genetic potential.
Although the genotype is established at
conception, parts of the genotype may not
be expressed until certain ages are reached.
20
HUMAN INTELLIGENCE
For example, there are a number of medi¬
cal conditions, such as Alzheimer’s disease,
that have some genetic basis, but are not
displayed until well past childhood.
The extent to which the genetic potential
is realized depends upon the extent to which
the environment encourages the develop¬
ment of intelligence. In modern developed
societies it appears that somewhere upwards
of 50% of the variance in IQ scores can
be statistically associated with genetic vari¬
ation. However, no one, ever, inherited a
test score in the same sense that they inher¬
ited the color of their eyes. People do dif¬
fer in the extent to which they have inher¬
ited brain mechanisms that allow them to
deal with their society in a way that pro¬
duces the mental capabilities required to
solve the problems proposed on a cogni¬
tive test. My apologies for a complicated
sentence, but I cannot think of any other
way to make this very important point.
Genetic capacities unfold throughout life. In
fact, the association of genetics with intelli¬
gence test scores is higher in old age than in
adolescence.23
The extent to which the genetic poten¬
tial is realized is determined by the physical
and social environment. Here is a striking
example.
If a pregnant woman abuses alcohol her
child may be born with fetal alcohol syn¬
drome, a serious form of mental retardation.
The likelihood that this physical condition
will occur depends on the social environ¬
ment. Fetal alcoholism does not incur in
societies that enforce abstinence. It can be a
substantial problem in societies where alco¬
hol is freely available as a recreational drug,
especially if the social stresses that lead to
alcohol abuse are present.
Social influences also act upon the devel¬
oping mind. At this point, though, we have
to shift from the concept of brain to the
concept of mind. Ultimately everything is
in the brain. Nevertheless, it is often useful
to distinguish between mental capabilities
that are determined by a person’s capacity to
23 McGue et al., 1993.
process information in the abstract and capa¬
bilities that are determined by the possession
of information, either about the world or
about how to solve problems in the world.
Information and problem-solving styles are
heavily influenced by the society within
which a person lives.
Arguably the most striking example is
literacy. The ability to read appears to
be associated with willingness to evaluate
abstract arguments and to take multiple per¬
spectives, a behavior that is tested in many
intelligence tests.24 These can be considered
direct influences of literacy upon intel¬
ligence. Literacy also has an indirect
influence, for it opens the door to formal
education, and education is, by definition,
a major avenue for passing on culturally
acquired knowledge. Literacy facilitates the
transmission of cultural knowledge, and
cultural knowledge allows us to behave
more intelligently.
Medicine offers an example. Modern
health workers wash their hands when
they move from patient to patient. Prior
to the mid nineteenth century this was
not seen as necessary. Does this make
today’s physicians more intelligent than the
physicians of George Washington’s day? In
an important sense, they are. Knowledge is
cognitive power.
There is a two-way interaction between
brain and mind. One of the major limita¬
tions on knowledge acquisition is the ability
to concentrate mental effort on a topic, in
the face of distractions. The ability to con¬
centrate depends on how well certain brain
structures work, primarily but not exclu¬
sively in the forebrain. It also depends upon
how efficiently information about the exter¬
nal world is coded inside the nervous system.
The coding systems people use depend very
much upon their experience with the topic
at hand.
Intelligence is a personal trait produced
by an interaction between genetic potential
and environmental support. Where do IQ
scores fit into this picture?
24 Cole, 2005; Wolf, 2007.
THE ISSUE OF INTELLIGENCE
21
1.4.3. The Measurement of Intelligence
The center portion of Figure 1.3 makes a
point that is obvious when you think about
it, but that is surprisingly often forgotten.
A person's intelligence test score is pro¬
duced by two things: the social decision to
construct an intelligence test in a particu¬
lar way and the examinee's ability to deal
with the test once it has been constructed.
Different societies might construct differ¬
ent tests, depending upon the mental capa¬
bilities that each society values. This does
not mean that tests are narrowly culture-
bound, for some mental qualities are seen
as vital in all societies. For instance, all soci¬
eties demand that their members learn their
native language. On the other hand, soci¬
eties may differ in the emphasis that they
place on other aspects of cognition.
Both these points were illustrated nicely
by an anecdote told me by Manuel de Juan-
Espinosa, a Spanish psychologist who has
studied conceptions of intelligence held by
the Fang, a society of mixed agricultural¬
ists and hunters in Equatorial Guinea. When
you ask people in Western society to list
the attributes of an intelligent person, you
generally get statements about the ability
to solve abstract problems and the ability
to comprehend and use language. Spatial
orientation, the ability to locate oneself in
the physical environment, is either not men¬
tioned or mentioned far down on the list.
When the topic is brought to their attention
people do agree that it is intelligent to be
able to find your way around the neighbor¬
hood. In our society, though, this is usually
not an important skill. Our intelligence tests
include only a few evaluations of spatial ori¬
entation.
When the Fang list the qualities of an
intelligent person they say, “Intelligent peo¬
ple do not get lost in the forest.” This does
not mean that the Fang devalue the sorts
of verbal skills that Westerners mention. In
fact, they stress verbal skills, in much the
same way that Westerners do. If the Fang
were to construct intelligence tests they
would include tests of verbal skills, as ours
do. The Fang might look into the matter
of spatial orientation in more detail than
we do.
IQ tests, and tests related to IQ tests, such
as the SAT and AFQT, are artifacts of the
cultures in which they arose. They test some
aspects of intelligence but not others. How¬
ever, the tests are not arbitrary.
IQ tests would not have survived, as arti¬
facts, unless test scores could be used as
(imperfect) predictors of what our society
sees as socially important behaviors, such as
academic and social achievement. Because
the test scores do meet this criterion, the
tests must either evaluate mental skills that
are used by the society or they must evalu¬
ate mental skills that are not used, in them¬
selves, but whose possession is highly corre¬
lated with the possession of skills that can
be used. That is, the tests might be like an
Army physical examination where the can¬
didate is required to do push-ups. Soldiers
are not going to do push-ups in combat,
but the ability to do push-ups is correlated
with the ability to move heavy objects (e.g.,
carrying artillery shells to a gun position),
which soldiers may have to do. The same
argument may apply to the mental gymnas¬
tics required to perform well on intelligence
tests.
While all human societies are not iden¬
tical, there is a core set of cognitive skills
that all societies rely upon. All societies
demand that their members learn to speak
the native language, be able to control atten¬
tion, and, by the standards of all other ani¬
mals in the world, remember events very
well. Therefore, an intelligence test that is
valid in one culture is unlikely to be entirely
invalid in another, although its validity may
be reduced.
Individual differences in cognitive skills
might, in principle, either be due to posses¬
sion of a great many special-purpose brain
mechanisms or be due to possession of
very general information-processing capa¬
bilities that can be applied to all men¬
tal challenges. Intermediate solutions are
possible; we might have a single general
processing capacity, augmented by special
processors. To the extent that the evolu¬
tion of our species has produced a general
22
HUMAN INTELLIGENCE
problem-solving brain, it does not matter
precisely how that brain is evaluated, for a
person's behavior in one cognitively chal¬
lenging situation will predict how well he or
she deals with other situations.
There is a surprising amount of evi¬
dence that evolution has taken the general
problem-solving solution.
1.4.4. The Uses of Intelligence
We now come to the right-hand section of
the diagram, dealing with the results of intel¬
ligence.
People use their cognitive abilities to
define their environments. Consider good
health practices. Sometimes very bright
people become alcoholics and/or crash cars,
both activities that can lead to brain dam¬
age. But, on the average, people with high
intelligence test scores do not do such things,
and so enjoy better health.25 Once again, we
see how intelligence and the environment
interact.
Cognitive abilities are not the only abil¬
ities that we have. Differences in behavior
are also produced by individual differences
in a variety of emotional-motivational traits.
These are lumped together loosely under
the title “personality.” It is not clear where
to draw the line between intelligence and
personality. For instance, conscientiousness,
the tendency to fulfill obligations to others
(including one’s employers) is usually con¬
sidered a personality trait. However, it is
possible to see conscientiousness as an off¬
shoot of intelligence. It makes sense to be
conscientious in fulfilling obligations to peo¬
ple who control resources that you want.
This is hardly a new observation. Machi-
avelli’s famous sixteenth-century discourse
on political behavior, The Prince, contained
cogent arguments for displaying certain
personality traits, including honesty and
conscientiousness, because such behavior
is in one’s enlightened self-interest. It is
intelligent to be thought trustworthy and
reliable.
25 For a typical study recent study, see Batty et al.,
2006.
We can make a distinction between cog¬
nition and personality by asking if we are
talking about whether a person can do or
will do a certain behavior. To the extent that
the answer is “can do” the mental acts con¬
trolling the behavior are part of cognition,
and hence individual differences in them are
part of intelligence. To the extent that the
answer is “will do” the mental control is part
of motivation, and individual differences are
part of personality.
Any particular action has to satisfy
both “can do” and “will do” requirements.
Although this is hardly a profound state¬
ment, it is surprising how often explana¬
tions of behavior focus on personality to the
exclusion of intelligence or vice versa.
1.4.5. The Results of Intelligence
Intelligence shows itself in two ways: by
test scores and by socially relevant behav¬
iors. Test scores are easy to analyze; socially
relevant behavior is hard to analyze. Nev¬
ertheless, socially relevant behaviors are far
more important than test scores.
There are statistical associations (cor¬
relations) between intelligence test scores
and measures of socially relevant behavior
including academic achievement, income,
health, and occupation of prestigious posi¬
tions in society.26 These associations are
facts. Arguing about why they occur is a
reasonable thing to do. As every student in
elementary courses on experimental design
is told, correlation does not imply causation.
Rather, the fact that intelligence test scores
and measures of socially relevant behaviors
are correlated suggests a number of possible
causes, all of which are worth investigation.
Performance on an intelligence test and
performance in socially relevant situations
might both depend upon the same cogni¬
tive processes. This is what most researchers
on intelligence believe. However, the sta¬
tistical associations, although not negligible,
are not compellingly large. They generally
range from a correlation coefficient of .2 to
26 Gottfredson, 1997; Herrnstein & Murray, 1994. See
the extensive discussion in Chapter 10.
THE ISSUE OF INTELLIGENCE
23
.5, and in a few situations correlations as high
as .8 have been reported. Just what these
statistics mean is explained in more detail in
Chapter 2. Temporarily, I ask readers to take
on faith the statement that the correlations
are high enough to indicate that test scores
and behaviors tap some common traits, but
low enough to indicate that the traits that
affect the tests and the behaviors are not
exactly the same.
It could be that the test scores themselves
are causing the behaviors but that the cog¬
nitive capabilities required by the tests are
not. This might happen in two ways. One is
that knowledge of a high or low test score
could cause other people to treat the exam¬
inee differentially. It has been claimed that
this is the case in education, because teach¬
ers may respond differently to students with
high or low IQ scores.27
Test scores will influence behavior if the
scores are used to decide who is offered
a chance to behave in a certain way. The
SAT, the ASVAB, and the Department of
Defense's Officers Qualifying Test (OQT)
are examples. Other things being equal, the
people with the highest test scores will be
allowed to enter a prestigious university, be
commissioned as an officer in the armed
services, and so forth. Subsequent behav¬
ior will be guided by the experience of hav¬
ing been in the university, been a commis¬
sioned officer, and so forth. On the one
hand, it can be argued that this is appro¬
priate. If a test correctly identifies people
who have the cognitive capability to, say,
benefit intellectually from an Ivy League
education, then the test is an appropriate
gatekeeper. On the other hand, it could
be argued that the benefits one gets from
being allowed to enter a certain social stra¬
tum depend very little on cognitive abili¬
ties. Colloquially, it’s not what you learn
27 Should teachers know their students’ IQ scores?
There are rational arguments for and against the
practice. If teachers know students’ IQ scores they
might offer special instruction to students they
think are bright, or they might give up on teach¬
ing low-scoring students. Looked at another way,
the information could be used diagnostically, help¬
ing teachers tailor instruction to the student.
Figure 1.4. People possess cognitive skills
(conceptual intelligence) to varying degrees.
Some of these skills are evaluated by the tests.
Of the skills evaluated by the tests some are
specific to the test situation, and others are
relevant to social behavior in general.
at Harvard (Yale, Stanford, Duke, Oxford,
or Cambridge) that determines your subse¬
quent success in life, it is who you meet in
college. Either of these mechanisms could
produce a correlation between test scores
and later success, on a between-institution
basis, because the test scores were used to
determine who got in, but not on a within-
institution basis, because the test scores did
not tap the abilities required to benefit from
the experience.
The two explanations are not mutually
exclusive. My own bet is that both factors
operate.
Figure 1.4 summarizes the argument.
Socially relevant behaviors are partly deter¬
mined by cognitive skills. Variations in
these behaviors are, conceptually, intelli¬
gence. Some of the skills that define concep¬
tual intelligence are reflected in test scores.
Other socially important cognitive skills that
are part of intelligence are not reflected in
test scores, and test scores also reflect test-
specific skills that are not related to socially
relevant behavior.
Lippmann was right that IQ scores do not
measure everything, Boring was right that
they do measure something.
Intelligence tests do a good job of evaluat¬
ing an examinee's ability to respond rapidly
to problems of varying degrees of complex¬
ity. In the following chapters I discuss evi¬
dence showing that IQ and similar tests are
statistically related to the cognitive abili¬
ties that underlie socially relevant behaviors,
ranging literally from becoming a judge to
24
HUMAN INTELLIGENCE
becoming a criminal. However, important
cognitive abilities required for socially rele¬
vant behavior are not tapped by the tests.
What might these be?
Any testing method that relies on the
conventional “Drop in from the Sky"
paradigm cannot evaluate abilities that
reveal themselves over a relatively long
period of time, such as the ability to plan, to
allocate time for extended courses of action,
or to integrate information from multiple
sources. It is difficult, if not impossible, to
tap these skills in a test that seldom takes
more than three hours and that consists of
unrelated problems that, individually, take
only a few minutes to solve. Overreliance
on conventional testing has greatly limited
modern research on intelligence.
1.4.6. Cause and Effect in Intelligence
Research
The arrows going back and forth in Fig¬
ure 1.3 highlight how hard it is to deter¬
mine cause and effect when studying intelli¬
gence. If everything were simple we would
place causal variables on the left, intelli¬
gence in the center, and the effects of hav¬
ing intelligence on the right. To some extent
this can be done. Genetics and a child’s
physical and social environment do pro¬
duce intelligence, and a person’s intelligence
does produce socially relevant behavior,
and thus alters the person’s environment.
The problem is with the feedback. Intel¬
ligence influences a person’s environment,
and feedback from the environment alters
intelligence.
Take the case of aging. As we grow
older two different processes influence intel¬
ligence. There is a decline in brain func¬
tion. At the same time, as people live they
acquire “wisdom," better and better knowl¬
edge about how the culture runs. There
are huge individual differences in both of
these processes. Some people remain cogni¬
tively fit until great age; others descend into
near-senility at little past fifty. Some peo¬
ple acquire wisdom as they pass through the
world; others only have experiences. In both
these situations intelligence appears to act
as both cause and effect. Other things being
equal, more intelligent adults are more likely
to maintain healthy lifestyles than less intel¬
ligent ones, and are more open to engaging
with, and thus extracting wisdom from, the
world about them.
The interaction between intelligence and
the environment has posed a major prob¬
lem for researchers. We know far less than
we need to know about the development of
cognitive power over the adult life span. It
is reasonably easy to measure cognition and
other psychological variables as long as peo¬
ple are in the educational system. Similarly,
it is relatively easy to obtain access to retired
people, through social and medical support
institutions, ranging from gray-haired hikers
pausing at an elder hostel to the patients in
a medical care facility. It is much harder to
obtain access to people in the working years,
for the simple reason that they are busy at
work and raising families.
This is a serious situation, for adult intel¬
ligence is not inert. It rises to meet environ¬
mental challenges.
1.5. Reaction Ranges and
the Challenge Hypothesis
Imagine a hypothetical person, Harry P., at
age fifty. We want to estimate Harry’s intel¬
ligence without actually measuring it. Here
is what we can do.
1. Our initial estimate is the average intel¬
ligence of a person at age fifty.
2. Collect facts about Harry’s physio¬
logical status, including genetic back¬
ground, medical history, present eating
and drinking habits, and so on. Use these
to compute a biological correction fac¬
tor to the initial estimate.
3. Collect facts about Harry’s social envi¬
ronment, including education, marital
status, hobbies, profession, and so forth.
Compute a social correction factor, and
add it in. The twice-adjusted figure is
our estimate.
The result will be an estimate of current
intelligence. We would not know how that
THE ISSUE OF INTELLIGENCE
25
Figure 1.5. The concept of reaction range. The ordinate represents
the quality of observable performance, and the abcissa represents
the extent to which the environment supports development of
cognitive skills. A person's level of cognitive performance is
determined by the combination of reaction range and quality of the
environment. Persons A, B, and C each have a unique potential for
cognitive performance, indicated by the three lines. Their actual
performance will be determined by the quality of the environment.
If the environment goes from very poor (at the far left of the figure}
to moderate (toward the middle), there will be considerable
improvement of performance. Going from moderate to good
environments (towards the right) does not result in a great deal of
improvement.
intelligence had been acquired. Estimation
alone does not explain the dynamics of intel¬
ligence.
A person’s genetic makeup does not pro¬
vide that person with a certain number of
“intelligence units/’ any more than it pro¬
vides a person with a certain number of
points on an IQ test. Genetic status pro¬
vides a reaction range , a range of levels of
intelligence. Environmental factors deter¬
mine where the person will operate within
that range.
Figure 1.5 displays three hypothetical
reaction ranges. Cognitive functioning is
shown on the ordinate (y-axis), the level
of favorableness of the environment on the
abscissa (x-axis). Three different reaction
ranges are shown. Where a person actu¬
ally functions is determined by the point at
which a vertical line drawn from the envi¬
ronment’s rating crosses the reaction range
line. Observable cognitive performance is
determined by the combination of genetic
reaction range and environmental quality.
The level of genetic inheritance cannot be
inferred from IQ scores unless environmen¬
tal quality is known, nor can environmen¬
tal quality be inferred from IQ scores unless
genetic inheritance is known.
Varieties of this theme will appear
throughout our discussion of intelligence.
For example, any difference in the cogni¬
tive performance of identical (monozygotic
or MZ) twins can be attributed to the envi¬
ronment because MZ twins, having identi¬
cal genotypes, will have identical genetically
determined reaction ranges.
The concept of reaction range applies to
the environment as well. This can be seen in
Figure 1.6, which is simply a “cleaned up”
version of Figure 1.5, in which three ver¬
tical lines have been drawn upward from
26
HUMAN INTELLIGENCE
140
Figure 1.6. Environmental reaction ranges. Suppose the
environmental quality is fixed (arbitrarily) at a quality of 10 for all
individuals. Differences in cognitive performance will occur due to
individual differences in genetic potential. However; if individual
A is placed in an environment of quality 15 and C in an
environment of quality 5, A will outperform C even though C has
greater genetic potential.
the environmental quality axis, each line
representing a different quality of environ¬
ment. Each of the lines intersects the differ¬
ent genetically determined reaction ranges
at different points, resulting in a variety of
cognitive performances, even though the
environment is identical for all individuals.
Figures 1.5 and 1.6 show the genetic reac¬
tion ranges as negatively accelerated curves
that rise steeply at first, and then flatten
out as they approach an asymptotic value.
This was an arbitrary choice, for both the
cognitive performance and environmental
quality axes are shown on arbitrary scales.
I chose to display negatively accelerated
curves because of my conjecture that, in
fact, attempts to improve cognition will
result in negatively accelerated curves. Why
do I make this conjecture?
There are a number of ways of produc¬
ing physical environments that greatly con¬
strain the development of intelligence. Pro¬
longed famine is one; infection of the brain
is another. Once these disaster states are
avoided, the incremental effects of improv¬
ing the physical environment are probably
rather small. The same thing is true for the
establishment of genetic potential. There
are many “catastrophic” genetic conditions,
where the presence of a single anomaly
greatly restricts cognitive development, but
if there is no anomaly genetic potential
seems to depend upon the combined effects
of a large number of genes, no one of which
contributes a great deal. These matters are
discussed in some detail in Chapters 8 and
9. There are lots of ways to restrict intelli¬
gence, but we know of relatively few ways
to expand it. This implies negatively accel¬
erated growth curves.
Having a reaction range does not mean
that you use it. Cognitive skills are acquired
by investing time and energy. Willingness to
invest depends upon one’s perception of the
likely outcome of the investment. I propose
the following conjecture.
The Challenge Hypothesis:
Intelligence is developed by engaging in
cognitively challenging activities. Envi¬
ronments vary in the extent to which they
support such challenges, and individuals
vary in the extent to which they seek them
out .
THE ISSUE OF INTELLIGENCE
27
The development of intelligence depends
upon the extent to which the individual
wishes, is allowed to, or is required to meet
environmental challenges.
Consider one of the linchpins of human
cognition - language. Every human being is
required to learn a spoken first language,
and all normal people do so. Only in the
last two hundred years have societies begun
to demand literacy, which is a much harder
skill to acquire. It is worth the effort. A liter¬
ate person has acquired a cognitive skill that
makes him or her more intelligent than an
illiterate one with an identical, but unreal¬
ized, biological potential for intelligence.28
The example expands. Literate societies
dominate the globe. They have developed
formal education systems that force fur¬
ther intellectual development. Mechanisms
of education such as the school, the news¬
paper, and the World Wide Web are differ¬
ent from, and far more challenging than, the
educational systems of the nineteenth cen¬
tury. As a result, today’s children are more
intelligent than children in the past. As a
reflection of this phenomenon, IQ scores
rose throughout the twentieth century.
Literacy is an example of a compulsory
challenge; individuals have to meet it or suf¬
fer severe consequences. In other cases there
may be options. Learning about statistics
and probability provides a good example.
On a population basis, relatively few peo¬
ple study statistics. Students who graduate
from, and understand, elementary statistics
courses can solve problems that Pascal and
Gauss could not solve. I do not think that the
typical student in a modern statistics class
has the biological potential for intelligence
that Pascal and Gauss had. There is a sense
in which they are more intelligent, for they
are more powerful problem solvers.
Robert Sternberg has identified three
ways a person has of responding to cogni¬
tive challenges.
1. Adapting : Changing your own cogni¬
tive behaviors to meet the challenge.
28 See Wolf (2007) for a very well-argued, extensive
expansion on this point.
Studying is a way of adapting, though
certainly not the only one.
2. Shaping : Changing the environment to
adapt it to your current capabilities.
If you have difficulty doing arithmetic,
buy a hand calculator.
3. Selecting : Finding a new environment
that does not present the challenge you
want to avoid. If you are a college stu¬
dent majoring in engineering, but math
classes are difficult for you, consider
switching majors.
Sternberg’s three strategies have differ¬
ent cognitive demands of their own. Both
adapting and shaping require a willingness
to engage with the environment. This is in
itself a reliable individual trait. Philip Acker¬
man, a professor at the Georgia Institute of
Technology, has conducted research show¬
ing that willingness to engage in intellectual
challenge is characteristic of the people who
hold the occupations and avocations that we
think of as requiring intelligence.29 Select¬
ing can be rational in some situations, but it
has the danger of becoming a way to avoid
intellectual challenge, and hence, to avoid
the development of one’s biological poten¬
tial for intelligence.
1.6. Intelligence Is Part of a System
Defining intelligence solely in terms of test
performance is an impoverished view. It
focuses our attention on explaining varia¬
tions in test scores, which are not impor¬
tant in themselves, at the expense of
studying individual variations in socially rel¬
evant behavior, which are important. How¬
ever, it would be foolish to disregard the
considerable amount of information about
intelligence that is incorporated in test
scores.
One of the major reasons for studying
intelligence is to understand how individual
differences in cognitive competence, intelli¬
gence in the broad sense, are related to indi¬
vidual differences in the display of socially
29 Ackerman, 1996; Ackerman & Beier, 2005.
28
HUMAN INTELLIGENCE
relevant behaviors. In practice; there are
two problems. Variations in socially relevant
performance are determined by noncogni-
tive as well as cognitive factors. Success (or
failure) is based on both “can do" and “will
do.” In addition, the display of socially rel¬
evant behaviors depends upon the opportu¬
nity to display them, which may be quite
beyond a person's control, no matter what
his or her personal characteristics are. Here
are two historical examples.
In the early twentieth century Joseph P.
Kennedy, an American financier, amassed
a considerable fortune. Subsequently three
of his sons - John, Robert, and Edward -
became United States Senators. In i960
John became President of the United States.
One can argue, probably correctly, that
Joseph Kennedy provided his sons the
genetic capability to become very intelli¬
gent. But his fortune certainly helped their
political careers, if only to relieve them of
the need to earn a living outside of public
service.
Now let us look at an example of restric¬
tion. The Declaration of Independence of
the United States, written in 1776, contains
the statement, “All men are created equal. ”
At that time women were disenfranchised,
and the slavery of Africans was condoned.
Over two hundred years later Condoleeza
Rice, an African American woman, who had
served in many high-level positions, includ¬
ing Secretary of State, observed that when
the Declaration was written, “They weren’t
talking about me." Dr. Rice’s impressive
achievements, which depended very largely
on her intelligence, would have been impos¬
sible in 1776.
At the start of the twenty-first century the
Darfur region of the Sudan was wracked by
drought, famine, and vicious ethnic warfare.
Children born in Darfur in the year 2003, no
matter what their genetic potential, did not
have good life prospects.
The point of these examples is that both
the causes and effects of intelligence are
embedded in a matrix of other variables.
This makes good research on intelligence
hard to do. The task is difficult, but it is
not impossible. While an ideal study may
be impossible to implement, a great deal
can be learned from less-than-ideal studies.
Progress can be made by investigating the
issues that can be studied, and hopefully by
holding down excessive interpretations and
conclusions where we do not have the right
evidence.
It is fairly easy to determine the corre¬
lation between test scores and other mea¬
sures of interest, such as grade point average
(GPA) and performance at work. Such stud¬
ies provide important data. However, they
bias our knowledge toward finding out the
role of intelligence in certain institutions,
such as the schools and the military, that,
while certainly important, are not the whole
of society. In addition, the study of bivari¬
ate correlations, in isolation, fails to stress
an important fact. Intelligence is just one of
the variables in the system defined by human
society. What does this mean?
A system is a set of interdependent vari¬
ables, in which each variable influences the
others. In closed systems the interdependence
is complete. The value of each variable is
completely determined by the other vari¬
ables in the system. In open systems some
variables are influenced by conditions out¬
side the system. The real-world systems
we study are always open. Therefore, it
is important to distinguish between system
variables, which exert measured reciprocal
influences on each other, and external vari¬
ables , which influence the system variables
but are not (to any great extent) influenced
by the system variables }°
To illustrate, consider a hypothetical
study of the roles of genetics, family
influence, and intelligence during primary
school, middle school, and high school. The
following relations hold:
1. A child’s intelligence on the first day
of school has been determined by
genetic inheritance and family environ¬
ment prior to entering school.
2. Intelligence on entering middle school is
determined by intelligence on entering
30 In economics the terms endogenous and exogenous
variables are used.
THE ISSUE OF INTELLIGENCE
29
? ? ?
Figure 1.7. A systems' view of the relationship between
intelligence, family environment, and the quality of schooling.
Intelligence and family environment are system variables. Genetic
heritage and quality of school are external variables. Unknown
sources that influence the system variables are indicated by a
question mark.
primary school, the quality of the pri¬
mary school, and family environment
during primary school, but not by the
environment prior to entry into primary
school.
3. A child’ s intelligence on entering high
school is determined by intelligence on
entering middle school, family environ¬
ment during middle school, and the
quality of the middle school.
4. At any time family environment is
determined by prior family environ¬
ment and by the child’s intelligence.
(People, including children, influence
their environment.)
Figure 1.7 diagrams the system. Intelli¬
gence and family environment are system
variables, because they influence each other.
Genetic potential and quality of schooling
are external variables, because they exert
influences on other measures but are not
influenced by them. If we can obtain mea¬
surements of all these variables, we can use
modern statistical methods to evaluate the
relative influences of each variable upon the
others.
Then we come to the external, unmea¬
sured variables. Figure 1.7 makes no pro¬
vision for extrafamilial influences on the
family environment, such as financial emer¬
gencies. Nor is there any provision for
extrafamilial and extra-educational influ¬
ences on intelligence, such as physical
injury. Therefore, we have to allow for the
“unknown unknowns,” the influence of
unmeasured external variables. These are
indicated by the “?” symbols in the figure.
While we cannot identify these variables,
modern statistical techniques do allow us
to estimate the size of their influence com¬
pared to the influence of the measured sys¬
tem variables on each other. Hopefully the
influence of the “unknown unknowns” will
be small; if it is large, analyses within the
system will account for only a small part of
what we need to know.
We can learn a great deal by compar¬
ing systems models to each other. The
intelligence-education system displayed in
Figure 1.7 treats genetic influences as a
one-shot effect - genetics influences intel¬
ligence prior to entering school - but has
no direct influence subsequently. In fact,
though, some genetic effects unfold over
time. For example, individual differences in
the rate at which connections are developed
in the forebrain during adolescence may
result in differences in the ability to control
impulsive behavior, which may influence
3°
HUMAN INTELLIGENCE
how much a student learns in middle school,
and so forth. These influences could be mod¬
eled by extending arrows from the genetic
inheritance box to the boxes represent¬
ing measurements of intelligence at each
time period, not just at the point of school
entrance. The extension is needed only if the
added arrows produce a system that (reli¬
ably) explains more of the variation in intel¬
ligence than did the simpler system.
System analyses of this sort allow
researchers to move far beyond arguments
over the meaning of a correlation between
IQ scores and just one other variable, in iso¬
lation from the system in which both occur.
Still, problems remain. Systems incorpo¬
rating human intelligence are so compli¬
cated that no one study can ever include
adequate measures of all relevant variables.
The size of the influence of unknown vari¬
ables can be evaluated, but what those vari¬
ables are and how they exert their influence
will remain unknown. Therefore, we cannot
explain all the causes and ramifications of
intelligence, all at once. We can identify and
analyze reasonably closed subsystems, deal¬
ing with a particular aspect of intelligence.
No one such study will tell us all about intel¬
ligence, but, taken together, they will tell us
quite a lot.
1.7. Summary and Prospectus
Individual differences in cognitive capaci¬
ties - intelligence, for short - are an impor¬
tant part of human variation. Intelligence
is partially tapped by intelligence tests (IQ)
and by other tests of cognitive achievement,
but other important cognitive abilities lie
outside of the tested realm. While intelli¬
gence does not completely determine suc¬
cess in life (or lack of the same), it certainly
contributes to it. In order to get a com¬
plete picture we have to consider personality
and motivational factors, and the extent to
which the social and physical environments
encourage some behaviors and discourage
others.
Intelligence has both multiple causes and
multiple consequences. In order to study
intelligence it is necessary to isolate rela¬
tively closed systems of variables. This poses
a challenge, because our ability to mea¬
sure some variables, and hence to study sys¬
tems involving them, is much greater than
our ability to measure other variables. Test
scores and measures of genetic variation are
much easier to obtain than measures of suc¬
cess in society, or measures of variation in
the physical and social environment. There¬
fore, we have to be vigilant against the error
of studying what is easy to analyze, at the
expense of missing amorphous but impor¬
tant effects. This tension will be reflected
throughout the discussions in this book.
Some typical tests will be described in
Chapter 2. Subsequent chapters deal with
the description, causes, and consequences of
intelligence. The book closes with a discus¬
sion of how intelligence is distributed across
our society and, finally, some speculations
about the development of research on intel¬
ligence in the future.
CHAPTER 2
The Tests
I agree with you that there is a natural
aristocracy among men. . . . the natural
aristocracy I consider as the most precious
gift of nature for the instruction, the trusts,
and the government of society . . .
Thomas Jefferson, letter to John
Adams, 18131
2.1. Introduction
Thomas Jefferson and John Adams, the men
who wrote in the Declaration of Indepen¬
dence that “all men are created equal,”
believed in a natural aristocracy! They
weren’t hypocrites. Jefferson and Adams
believed that all men had equal rights, not
that they had equal talents. Adams’s life¬
long support of his alma mater, Harvard
University, and Jefferson’s investment in the
University of Virginia showed how strongly
both men supported the development of
the natural aristocracy. However, neither
Jefferson nor Adams said how the natural
1 Quoted in Lemann, 1999.
aristocrats were to be identified. Today we
partially rely on tests to do this.
We all have an intuitive feeling about
what a cognitive test is, for you are unlikely
to grow up in post-industrial society without
taking one. This chapter expands on intu¬
itive notions by describing some of the typi¬
cal tests, chosen to represent different types,
and by commenting on the aspects of intel¬
ligence that they do, or do not, evaluate.
Most textbooks on testing stress the dis¬
tinction between individual testing and test¬
ing people in groups. An equally important
distinction is between tests composed of bat¬
teries of subtests and single-format tests.
Test batteries contain sub tests that eval¬
uate different aspects of cognition. For
instance, many test batteries contain, as sub¬
tests, vocabulary tests and tests of simple
arithmetic. Overall scores are constructed
by combining subtest scores. The way the
widely used Wechsler test batteries are
scored is illustrative. The Wechsler batter¬
ies contain subtests involving verbal and
nonverbal material. Subset scores are com¬
bined to obtain verbal IQ (VIQ) and perfor¬
mance IQ (PIQ) scores, and then verbal and
31
32
HUMAN INTELLIGENCE
performance scores are combined into a full-
scale IQ (FSIQ) score.
The SAT, a college admission test that
will be familiar to many readers, is a battery
of subtests that evaluate different aspects of
verbal and mathematical reasoning. We will
look at this test in more detail in section 2.3.1
of this chapter.
Single-format tests are tests in which all
questions are of the same type. Thus a
single-format test looks very much like one
of the subtests of a test battery, used alone.
For instance, vocabulary tests are sometimes
used alone, to get a rough indication of a per¬
son's general cognitive power. Obviously a
single format test does not evaluate as many
cognitive talents as a test battery does. How¬
ever, if scores on a single format test are
highly correlated with scores on the test bat¬
tery as a whole (as vocabulary scores and
VIQ scores are for the Wechsler tests), an
estimate of a person’s intelligence can be
made, with some loss of accuracy and great
reduction in cost, by using a single-format
test that takes a few minutes instead of a
test battery lasting several hours.
There is no board that certifies a test as
an “intelligence test.” Indeed, only a few
of the many cognitive tests on the market
are specifically labeled tests of intelligence.
Why? The answer to this question requires a
look at different tests, and a brief excursion
into what might be called “political seman¬
tics.” The case of the SAT provides a good
example.
The SAT evaluates cognition. However,
the Educational Testing Service (ETS), the
company that constructs the SAT, never
calls it an intelligence test. ETS says that the
test evaluates the extent to which a person
who has an American high school educa¬
tion is prepared to be a college undergrad¬
uate. This sounds very much like Binet's
goal for the very first intelligence test, which
was designed to assess children’s readiness
for the French public education system.2
So why isn’t the SAT an intelligence test?
In fact, some researchers treat the SAT as
virtually synonymous with a measure of
2 Binet & Simon, 1905.
intelligence, no matter what ETS says.3 The
same thing is true of several other cognitive
tests. Knowledgeable people call them intel¬
ligence tests, but their publishers do not.
What we have here is a problem of defini¬
tions, tempered by some political controver¬
sies. Historically, and contemporaneously,
certain tests have been called intelligence
tests. These include Terman’s translation of
Binet’s tests, the Stanford-Binet test, and
David Wechsler's Wechsler scales, to be
described later. Both are battery-type tests
intended to summarize the results of an eval¬
uation of a wide range of cognitive skills,
and to be appropriate for many different
populations.4
These individually administered, battery-
type tests did not fill two important niches.
They were too expensive for mass personnel
screening programs, such as military recruit¬
ment. Therefore, group-administered tests,
such as the Army Alpha and the SAT, were
developed. There was also a need for spe¬
cialized tests, to be used when the examiner
was interested in only a limited range of cog¬
nitive abilities or when the test was intended
for use in a particular population.
Some comparisons involving the SAT
illustrate all these issues. The SAT is a group
test, while the Wechsler intelligence tests
are individually administered. It costs much
less to evaluate a person using the SAT than
it does using the Wechsler. The Wechsler
scales cover a greater range of cognitive
skills than the SAT does. For instance, the
Wechsler scales include sub tests to evalu¬
ate visual-spatial reasoning. This is a reason¬
able thing to do, for visual-spatial reasoning
is certainly part of human cognition. The
SAT does not contain visual-spatial reason¬
ing tests, on the grounds that this ability is
not required throughout the college curricu¬
lum, while verbal and logical reasoning abil¬
ities are required everywhere.5
3 E.g., Jackson & Rushton, 2006.
4 Matarazzo, 1972; Wechsler, 1975.
5 Perhaps it should. A case can be made for consid¬
ering visual-spatial reasoning as an ability required
in majors such as art, architecture, and engineer¬
ing. See Humphreys & Lubinski, 1996, for a more
extended discussion.
THE TESTS
33
Now consider the SAT compared to
the Armed Services Vocational Aptitude
Battery (ASVAB). Both the SAT and the
ASVAB contain subtests evaluating lan¬
guage skills. The questions on the SAT are
harder than comparable questions on the
ASVAB, on the grounds that a higher level
of language skill is required to succeed as a
college student than to succeed as a military
enlisted person. Both tests evaluate linguis¬
tic skills, but in different ranges.
Here is a more extreme illustration of
tests intended to evaluate different ranges of
intelligence. The Mini Mental Status Exam¬
ination is a test widely used in medical prac¬
tice to determine if a person is seriously cog¬
nitively impaired. It contains questions like
“What day is it?” and “Where are you?” At
the other end of the scale, the Miller Analo¬
gies test, intended to screen entrants to grad¬
uate school, contains some very hard analog¬
ical reasoning questions.
Given this variety of uses, it makes sense
to describe a test by the use for which it is
intended, instead of using the omnibus term
“intelligence test.”
A second reason for avoiding the words
“intelligence test” is dictated more by pub¬
lic relations than by a desire to provide an
accurate description. The very word intel¬
ligence has a negative connotation to some
people. Some social commentators associate
the term with elitism and even racism. The
reasons for this are discussed in Chapters 8
and 11. As a result some test producers deny
that they are evaluating intelligence, resort¬
ing to terms like “ability” and “aptitude,"
both to avoid controversy and to increase
marketability.
While the proliferation of test names is
understandable, it has resulted in a good deal
of inconsistency. In 2006 the authors of an
article in the journal Intelligence extracted
a score from the SAT that they referred
to as an index of general intelligence.6
Two issues later in the same journal, other
authors referred to the SAT as a test of
scholastic achievement, and tried to pre¬
dict SAT scores using scores on a nonverbal,
6 Jackson & Rushton, 2006.
academic-content-free test called the Raven
Advanced Progressive Matrices (RAPM).
The RAPM did pretty well as a predictor
(correlation of .39), but a vocabulary test did
substantially better (correlation of .69}. 7
This example shows, in microcosm, what
the situation is. On the one hand, virtu¬
ally all tests of cognitive abilities correlate
positively with each other. This shows that
there is a tendency for cognitive skills to vary
together, across people. If you have a high
level of one cognitive skill, you probably do
not have a low level on another. On the
other hand, the associations between cog¬
nitive skills are far less than perfect, so there
is a case for specialized tests in appropriate
situations.
With these general remarks aside, let us
look at some of the tests, as they could be
described by data available in the 2007-09
period. New tests and revisions are pub¬
lished frequently, but they all are designed
on the same basic principles.
2.2. A Description of Individually
Administered Test Batteries
Individual testing is generally appropriate if
there is concern about the mental capabili¬
ties of the examinee. For instance, in educa¬
tion a child might be examined for special
placement in a remedial (“special”) educa¬
tion program or, less frequently, for admis¬
sion to a program for gifted children. Tests
used for this purpose include the Stanford-
Binet, a much-updated version of Terman’s
translation of Binet’s tests, and the Wechsler
Intelligence Scale for Children (WISC).
Adults are tested individually as part of
a clinical assessment for a variety of behav¬
ioral problems. The judicial system some¬
times uses intelligence testing to determine
whether or not the examinee is sufficiently
intelligent to be held criminally responsi¬
ble for his or her (illegal) actions. Individ¬
ual testing is also used in order to assess an
adult’s mental status following injury to the
brain.
7 Rohde & Thomson, 2007.
34
HUMAN INTELLIGENCE
Intelligence tests are also widely used in
research on individual differences in cog¬
nition. Researchers tend to prefer the less
expensive group-administered tests, simply
to save money. In some situations, however,
individual testing is appropriate. As an alter¬
native, it is often possible to administer some
of the subtests of an individual test on a
group basis, for a special purpose, such as
the evaluation of vocabulary.
2.2.1. The Wechsler Tests: The Adult
Intelligence Scale (WA1S-IV) and the
Wechsler Intelligence Scale for Children
(WISC-IV)
The Wechsler tests for adults and children
(the WAIS and the WISC) are by far the
best-known individual tests. They are regu¬
larly revised, as are other intelligence tests.
Early versions of the WAIS tests provided
three scores: a verbal IQ score (VIQ) cal¬
culated from scores on subtests involving
language, a Performance IQ (PIQ) score
based on scores from sub tests that did not
involve language, and a Full Scale IQ (FSIQ)
based on a combination of the VIQ and PIQ
scores. A memory scale could also be
calculated. These scales, which are based
on a pragmatic division of cognitive skills
into verbal and nonverbal skills, are often
reported in all but the most recent literature.
The current (WAIS-IV) test has a some¬
what different structure, which reflects
recent theoretical developments, and in par¬
ticular the importance of individual differ¬
ences in the speed of cognitive process¬
ing and in the ability to keep information
in immediate memory while working on a
problem. The sub tests are grouped into tests
involving verbal comprehension, perceptual
(visual) reasoning, immediate (“working”)
memory, and speed of processing. An index
is calculated for each of the groups. The Full
Scale IQ score is calculated by combining
the indexes. Within each of the groups there
are core subtests, which are used to calcu¬
late the appropriate index, and supplemen¬
tal tests that can be utilized if the exam¬
iner wishes to probe the examinee’s abilities
within a particular area.
Table 2.1 lists the core tests used within
each group. It also includes brief descrip¬
tions of the cognitive skills required to com¬
plete each test.
Like the Stanford-Binet, the Wechsler
tests represent a pragmatic approach to
intelligence. They incorporate a widely
accepted distinction between verbal and
nonverbal reasoning. Otherwise the tests
and subtests have evolved, and theories
of intelligence have been induced from
research on them, rather than psychologists
having used a theory to generate the tests.
2.2.2. Two Individual Tests Motivated by
a Psychological Theory
The pragmatic approach to theory exempli¬
fied by the early Wechsler tests contrasts
with a more theory-based approach taken
in the development of two other widely
used tests: the Kaufmann intelligence tests
(one for adults and one for children) and
the Woodcock-Johnson test, which is used
largely for adult self-assessment as an aid in
career planning. These tests are based on a
theoretical distinction between crystallized
and fluid intelligence, first articulated by
Raymond Cattell and developed further by
his colleague John Horn.8
Cattell and Horn distinguished between
solving problems by applying previously
learned knowledge and/or problem-solving
methods ( crystallized intelligence, Gc) and
solving problems by applying general rea¬
soning methods to figure out a solution
to an unfamiliar problem ( fluid intelli¬
gence, G/). This echoes Juan Huarte de
San Juan’s sixteenth -century distinction
between problem solving based on mem¬
ory or imagination,9 and also resembles
Spearman's distinction between inductive
and deductive reasoning. Cattell and Horn
8 Cattell, 1971, 1987; Horn, 1985; Horn & Noll, 1994.
9 Cattell and Horn's ideas were developed indepen¬
dently of Huarte’s work. Huarte's book on intelli¬
gence was first translated into English in 1594, but it
appears to have had little influence on contempo¬
rary English-speaking psychologists. His ideas were
reintroduced to us by Spanish psychologists in the
1980s.
THE TESTS
35
Table 2.1. The subscales of the Wechsler Adult Intelligence Scale-TV (supplemental
scales are indicated)
Name of Test Content Cognitive Skill Evaluated
Tests used to compute the verbal comprehension scale
Vocabulary Word definitions required
Similarities Given names of objects, explain how
they are alike
Information Questions based on general cultural
knowledge
Comprehension Oral questions requiring solutions to
(supplemental) everyday problems
Tests used to compute the working memory scale
Digit Span Repeat a series of digits either forward or
backward
Arithmetic
Simple arithmetic problems are to be
solved “in the head.”
Letter-number
sequencing
(supplemental)
Examinee hears a mixed sequence of
letters and numbers, and repeats them
back separately in their normal sequence
Tests used to compute the perceptual reasoning scale
Block design Assemble a set of colored cubes to form a
specified design
Matrix reasoning
Visual puzzles
Picture completion
(supplemental)
Recognize a pattern of changes in visual
figures
Choose three of six pieces to make a
specified design
An incomplete picture of a common
object is shown. The task is say what is
missing
Figure weights A balance scale is shown, with some
(supplemental) objects in it. The task is to choose other
objects to balance the scale. Object
weights are to be inferred from examples
of balanced scales
Tests used to compute the processing speed index
Symbol search Indicate whether or not designated target
symbols appear in a group of symbols
Vocabulary
Abstract reasoning. Detection of
similarities
Knowledge of culture
Verbal comprehension, knowledge of
culture
Short-term storage of information
Manipulation of information in
immediate memory
Manipulation of information in
immediate memory
Nonverbal reasoning. Visualizing
object movements
Pattern recognition and application
Visualizing movement of objects.
Visual immediate memory
Nonverbal reasoning. General
knowledge
Nonverbal reasoning. The examinee is
forced to deal with
an unusual task
Speed of visual processing. Short-term
memory
Coding
Cancellation
(supplemental)
The examinee is shown an arbitrary
pairing of marks and numbers. The
examinee is then shown a set of marks
and must list the associated numbers
The examiner describes values of
attributes of objects, e.g., shape and
color. The examinee must indicate which
shapes in a set have the required
combination of attribute values
Paired associates learning. Speed of
visual processing
Speed of visual processing. Also, the
ability to disregard distracters, such as
a shape that has one of the required
attribute values (e.g., a red circle when
the target shape is a red triangle)
36 HUMAN INTELLIGENCE
Table 2.2. The subtests of the Kaufman Adult Intelligence Test
Name of Test Description of Activity0
Crystallized (Gc) tests
Definitions
A sentence is given with a missing word. Some letters of the missing word
are given. The task is to find the missing word. Example: The jockey was
riding a _ _0_S_
Auditory comprehension
The examinee listens to a brief recording, and then answers questions that
require either recall of information in the recording or drawing an
inference from the message.
Double meanings
The examinee is given two sets of word clues. The task is to find a word
that has two meanings, associated with each of them. Example: Clues:
“Sport & Champion,” “Music & Singing.” Answer: Record.
Famous people
(used for extended testing)
The examinee is shown a photograph of a famous person and a brief
statement about that person. The task is to name the person.
Fluid (Gf) tests
Rebus
The examinee is shown a “sentence” made up of pictures that stand for
words. The task is to “read” the sentence. Example: Show cartoon pictures
of an eye, a tomato soup can, and a fly. The sentence is “I can fly.”
Mystery codes
The examinee is shown a set of pictures, each with a “code meaning,” e.g.,
a large triangle representing “woman” and a black dot representing “new.”
The examinee is then shown a triangle with a black dot inside. The task is
to determine what the meaning of the new symbol is (“girl”).
Logical steps
The examinee hears or reads a set of premises, and then must deduce the
answer to a question. Example: “At the Round Table, Gawain is to
Arthur’s left and Launcelot is between Gawain and Arthur. Who is to
Launcelot’s left?”
Memory for block designs
(used for extended testing)
The examinee is shown a pattern composed of colored blocks. The
examinee must then recreate the pattern from memory, using blocks
and a form board.
a The examples are not of actual test items, in order to maintain the confidentiality of the test.
deserve credit for carrying this idea consid¬
erably further than their predecessors did.
Table 2.2 presents the sub tests of the
Kauffman Adult Intelligence Test (KAIT),
grouped into sub tests used to generate the
Gc and Gf scores. As in the case of the
WAIS, a “full-scale” score can also be gener¬
ated. The Woodcock-Johnson test also gen¬
erates separate scales for Gc and Gf, along
with a full-scale IQ score.
2.2.3. A Test Motivated by a Theory
of How the Brain Works
The Gc/Gf model and the distinction
between verbal and nonverbal reasoning are
models of how the mind works, for there is
no direct connection to brain mechanisms.
(In Chapter 7 we look at how such con¬
nections can be made.) The next test to
be described, the Cognitive Assessment Sys¬
tem (CAS), was motivated by a theory of
how the brain works. The theory and related
tests were developed by J. P. Das, a pro¬
fessor at the University of Alberta, and his
colleagues.10
Shortly after World War II, Das, a native
of India, had the opportunity to work in
the then Soviet Union, with the noted
neuropsychologist A. R. Luria. Luria had
10 Das, Naglieri, & Kirby, 1994.
THE TESTS
37
Figure 2.1. The four major lobes of the cerebral
cortex. Copyright Matthew Holt, used by
permission of the Washington University of
St. Louis Internet Stroke Center.
examined a very large number of brain
injury cases. He was struck by the fact
that damage to different areas of the brain
produced quite different behavioral deficits.
Luria concluded that the four major areas of
the cerebral cortex, the frontal, temporal,
parietal, and occipital lobes (see Figure 2.1),
are specialized for different types of infor¬
mation processing.11
It had already been determined that the
occipital lobe is largely concerned with the
mechanics of vision. Luria believed that
the temporal lobe is concerned with pro¬
cessing information in series (e.g., compre¬
hending a sentence as it is spoken), that the
parietal lobe is concerned with parallel pro¬
cessing, as when one notices the symmetry
of a figure, and that the frontal lobe is con¬
cerned with planning actions and with the
control of attention.
Luria was mainly concerned with the
results of substantial injuries to the brain.
Das extended Luria’s ideas to the anal¬
ysis of normal behavior by healthy chil¬
dren. He reasoned that if different areas
of the brain carry out planning, attention
control, and serial and simultaneous func¬
tions, then individual differences in behav¬
ior within the normal range should reflect
individual differences in the development
of each of the four brain areas. Das coined
the acronym PASS, for Planning, Attention,
Serial, and Simultaneous (PASS) model, and
constructed a number of tests to evaluate
each function. Das’s colleague Jack Naglieri
subsequently developed a comprehensive
test battery, the cognitive assessment sys¬
tem (CAS), incorporating subtests for the
four dimensions of the PASS theory.12
Table 2.3 shows the subtests used in the
CAS. The score on each subtest is based on
both the speed and accuracy with which a
child answers questions. An effort was made
to construct test items that were as nearly as
possible void of cultural content. This, and
the emphasis on the speed with which sim¬
ple tasks are done, is consistent with Das’s
and Naglieri 's desire to evaluate the effec¬
tiveness of brain processes rather than men¬
tal competencies tied to cultural knowledge.
Although Das was inspired by Luria’s
neuropsychological work, it is not clear that
the CAS scores map onto discrete brain
functions, for our understanding of the brain
is greater than it was in Luria’s day. The
PASS model is stated in terms of cognitive
behaviors: planning, attending, and process¬
ing information in either a parallel or serial
manner. These are characteristics of the
mind as an information-processing device,
not characteristics of the brain as a physical
device. The mapping to brain locations is
not quite what Luria’s theory predicts. For
instance, tasks that involve planning draw
heavily upon both frontal lobe and parietal
lobe mechanisms. 15 Also, the Das-Naglieri
assignment of tasks to the four PASS func¬
tions seems (to me) to be sometimes insight¬
ful and sometimes arbitrary. As an example
of a good assignment of task to function, the
planning connections task is a variant of a
“trail making” task, which is widely used as
an indication of damage to the frontal cor¬
tex. However, it is less clear that one should
regard the matching numbers and planned
codes tasks as examples of planning.
In the last analysis, though, the proof of
the pudding is in the eating. There are two
questions that we have to ask about any
proposed battery of tests. First, do the test
11 Luria, 1962/1980.
12 Naglieri, 2005.
13 Jung & Haier, 2007.
38 HUMAN INTELLIGENCE
Table 2.3. The subtests of the Cognitive Assessment System
Tests of planning
Matching numbers
Planned codes
Planned connections
Tests of attention
Expressive attention
Receptive attention
The examinee is shown rows of eight numbers. Two of the numbers
are identical. The task is to underline the identical numbers. Several
pages of rows are shown. The measure is the time needed to
complete the task.
The examinee is shown a set of letters and a code form for each; e.g,
A = OX, B = XX. The examinee then examines a number of rows of
letters and replaces them with their code form.
The examinee is shown numbers arranged haphazardly on a page.
The task is to draw a line (“trail”) starting with the lowest number
and proceeding in order to the highest. In a harder problem, numbers
and letters are intermixed, e.g., 1-A-2-B
This is a variety of the Stroop task, in which a person is shown words
in different-colored ink and must call out the color of the ink. The
words are color names, as in BLUE printed in red ink.
The examinee is presented with pairs of letters, e.g., AB, Aa, AA. The
task is to identify which ones have the same name.
Simultaneous processing tasks
Matrix task This is a progressive matrix task, similar to the matrix task on the
WAIS-R. It evaluates nonverbal reasoning.
Verbal-spatial relations
Figure memory
The examinee is shown six simple pictures and asked a question. The
examinee indicates which picture is the best answer to the question.
The examinee is shown a geometric pattern, and then must find it
embedded in a larger pattern.
Successive processing tasks
Word series The examiner speaks an arbitrary string of words. The examinee
repeats the words in order.
Sentence repetition The examiner speaks a syntactically correct sentence, which is
composed of color words and has no obvious semantic content, e.g.,
THE BROWN GREENS THE YELLOW. The examinee repeats the
sentence.
Sentence questions
The examiner speaks a sentence as in sentence repetition, then asks a
question about it, e.g, WHO GREENS THE YELLOW?
scores predict real-world cognitive behav¬
ior? Second, if the test is based on a the¬
ory, do the patterns of scores on the sub¬
tests behave in a manner predicted by the
theory? In the case of the CAS, are sub¬
tests within the Planning, Attention, Serial,
and Successive groups more highly corre¬
lated with other subtests within the same
group than they are with subtests in other
groups? We postpone the answers to these
questions, for the CAS and the other battery
tests, until our discussion of psychometric
theory, in Chapter 4.
2.2.4. ^ Commentary on IndiinduaUy
Administered Tests
Individual testing offers two advantages over
group testing; contingent item presentation
and supplementary reporting.
THE TESTS
39
Contingent item presentation simply
means that the next question that the exam¬
iner asks can be chosen on the basis of the
answers that the examinee has already given.
This is efficient and sensible, because the
examiner avoids trying to find out some¬
thing about the examinee that the examiner
already knows. If you are evaluating a per¬
son’s ability to do arithmetic and you have
already seen the person multiply 24 by 72 “in
the head,” there is not much point in asking
if he or she can add 32 and 18! In an examina¬
tion of a person's knowledge of US history, if
someone has already told you that Franklin
Pierce was president of the US after Millard
Fillmore, there is not much point in prob¬
ing to find out if they know who followed
Franklin Roosevelt. In addition to its effi¬
ciency, contingent item presentation makes
the examination more like an interview than
an examination, for the question-response-
next question sequence follows the rules
of normal conversation. Wechsler himself
referred to his scales as providing a struc¬
tured interview for the display of cognitive
skills.14
Contingent item presentation can also
be obtained by computer-controlled test¬
ing, a procedure that is becoming increas¬
ingly common. Computer-controlled testing
assumes that the examinee has basic com¬
puter skills and is motivated to cooperate.
Both assumptions are probably correct for
healthy adults and older children, in indus¬
trial and post-industrial societies. They are
suspect for young children, people who are
infirm (e.g., the extreme elderly), and out¬
side the post-industrial world.
Examiners typically write reports describ¬
ing how the examinee reacted during an
individual testing session. These comments
can be very helpful. For example, if intelli¬
gence is being examined to determine recov¬
ery from a brain injury, the examinee’s
behavior during testing may be as important
as the test score. However, there are two
qualifications. One is that the observations
and interpretations made during testing are
only as good as the observer, which raises
14 Wechsler, 1975.
the issue of variations in skill across examin¬
ers, as well as examinees. The other is that
both the cost and the difficulty of scoring
the examiner’s subjective impressions make
it difficult to incorporate such observations
into research programs.
The costs of individual intelligence exam¬
inations are in the $200-350 range.15 Some¬
times such costs are clearly justified. Sup¬
pose the purpose of the examination is to
decide whether or not a student should be
placed in a special (remedial) education pro¬
gram. The cost of supporting a student in
an American special education program is
about $17,000, more than twice the cost of
supporting a student in a regular program.16
Keeping a student in a regular classroom
when he or she needs special education can
be a disaster, both for the student and for the
teachers who must deal with a person who
cannot keep pace with his or her classmates.
In such a case individual testing is worth it,
from the viewpoint both of the institution
and of the person being examined.
2.3. Group- Administered Test Batteries
Individual examinations are not economi¬
cally feasible in many personnel selection
systems. Consider some numbers. In 2005,
US high schools graduated approximately
three million people. Slightly under half
of them took the SAT, the most widely
used college entrance examination.17 The
US military screens over 100,000 potential
recruits a year. A commercial testing com¬
pany that specializes in screening applicants
for hourly-wage positions uses computer¬
ized testing to screen an average of 40,000
people each workday. In such situations
individual examinations are not an option.
Group tests are needed.
15 A fee of $250 was quoted on the website of the
State University of New York, Stony Brook, Center
for Psychological Services, January 2010.
16 This figure is based on the National Education Asso¬
ciation's website, www.nea.org/specialed/index
.html, 14 February, 2007.
17 National Council of Educational Statistics and
College Board figures for 2005.
4o
HUMAN INTELLIGENCE
Table 2.4. The Otis-Lennon Test of School Abilities; selected subtests
Scab or Subtest Description
Verbal tests
Comprehension
Antonyms
Synonyms
Sentence completion
Verbal reasoning tests
Classification
Analogies
Arithmetic reasoning with
verbal problems
Nonverbal tests
Figural reasoning
Figural analogies
Matrix patterns
Given a base word, select from a list of target words the word
most nearly opposite in meaning.
Same as antonyms, except that the word to be selected should
have the same meaning as the base word.
Select word needed to complete an incomplete sentence.
Given a set of words, identify those that are in the same class.
Given two words as a base, and a target word, identify the
relationship between the two base words and use it to select a
word that has the same relation to the target word. Example:
White is to black as up is to (sideways, below, down, behind).
Word problems requiring simple computations.
An analogy test similar to verbal analogies, except that the
relations are between parts of geometric figures.
A matrix test using geometric figures (see the discussion of
progressive matrix tests in section 1.2).
Quantitative reasoning
Number series Given a series of numbers, where each number bears some
relation to the proceeding ones, identify the next number in the
sequence.
Number matrix Matrix problems similar to those used for figures, but
constructed using relations between numbers (see text below).
Note: The Otis-Lennon Test formats vary with grade level, the examples here have been chosen to
illustrate the nature of the test.
2.3.1. The Otis-Lennon Test
Most group-administered, battery-type
intelligence tests carry over the pragmatic
distinction between verbal and nonverbal
skills that characterizes the Wechsler and
Stanford-Binet tests. Tests of quantitative
skills are often added. In spite of the
similarity between these batteries and the
individual tests, test developers seldom
use the term intelligence. Instead the tests
are described as tests of “reasoning” or
“scholastic ability.”
A good example is the Otis-Lennon Test
of School Ability . This is a descendent of
Arthur Otis's Otis Group Intelligence Scab ,
first published in 1918, with the latest revi¬
sion in 2009. Otis, who had studied with Ter-
man, realized that most of the items on tests
like the Stanford-Binet could be written in
a form suitable for group administration.18
Accordingly he designed such tests for use
in the schools.
18 Robb, Bernardoni & Johnson, 1972.
THE TESTS
41
Table 2.5. The scales and subtests of the
Cognitive Abilities Test-3. (When the test
is administered in classrooms, three class
sessions are required, as indicated in the
table)
Session 1:
Verbal Battery
Test 1: Verbal Classification
Test 2: Sentence Completion
Test 3: Verbal Analogies
Session 2:
Quantitative Battery
Test 4: Quantitative Relations
Test 5: Number Series
Test 6: Equation Building
Session 3:
Nonverbal Battery
Test 7: Figure Classification
Test 8: Figure Analogies
Test 9: Figure Analysis
The Otis-Lennon test is actually a test¬
ing program. Below the fourth grade chil¬
dren are given individual tests. Group-
administered, age-appropriate group tests
are provided for grades five through twelve.
Table 2.4 shows some of the subtests used.
The test battery is divided into verbal and
nonverbal sections. Within the verbal sec¬
tion a further distinction is made between
verbal comprehension and verbal reason¬
ing. The nonverbal section is divided into
tests intended to evaluate “figural reasoning”
(relationships between visual patterns) and
quantitative reasoning. The test as a whole
takes about three hours, but it is typically
taken in several sessions.
2.3.2. The Cognitive Abilities Test
A similar approach to testing is seen in
the Cognitive Abilities Test (CAT-3, for
the third revision), developed by David
Lohman, Robert Thorndike, and Elizabeth
Hagen, and marketed through Riverside
Press. The publishers claim that this test
is the most widely used intelligence test in
the United Kingdom. It has figured promi¬
nently in important studies of the rela¬
tion between intelligence and educational
accomplishment.
Table 2.5 shows the sections and sub¬
tests of the CAT-3. There is a substantial
commonality between the Otis-Lennon and
CAT-3. This is hardly surprising, as the
tests were developed for the same purpose,
for the same audience, using a pragmatic
approach of building content on the content
of previous tests, rather than conforming to
a theory of intelligence.
2.3.3. The SAT
We now look at one of the best-known,
most-discussed tests used in education, the
SAT, which is widely used as part of the col¬
lege admission process in the United States.
The SAT is a distant descendent of a
college admissions test developed in the
1920s by Carl Brigham, a psychologist who
had worked on construction of the Army
Alpha Test. Brigham observed that the
Army Alpha was easy for candidates who
had some college education. He concluded
that a test similar to the Army Alpha could
be developed as a college entrance test, but
that it would have to be made more difficult.
Brigham produced a test that was used on
a trial basis by several universities during the
1920s and 1930s. In 1933 Harvard's president,
James Bryant Conant, became interested.
The tests that Harvard used at the time
emphasized the content of courses available
in a relatively small group of expensive East¬
ern preparatory schools. Conant knew that
this gave an advantage to children from fam¬
ilies who could afford to send their chil¬
dren to these exclusive schools. Accordingly,
Conant asked one of his faculty members,
Henry Chauncey, to look into developing
tests that would identify gifted students who
had not attended private schools.
Chauncey himself was a graduate of a
private school, although he was not excep¬
tionally rich. Nevertheless, in athletic par¬
lance, he took the ball and ran with it. He
was impressed by Brigham’s test, but could
not adopt it due to the disruptions caused
by the Depression and World War II. After
the war ended a US presidential commis¬
sion concluded that there was a need to
open up college admissions procedures to a
wider range of applicants. The commission's
conclusion mirrored the concerns Conant
42
HUMAN INTELLIGENCE
Table 2.6. The subtests of the SAT-I
Test Section
Item Type
Comment
Verbal comprehension
All questions are multiple choice.
Reading comprehension
Read short passage and answer
multiple-choice questions
The questions deal with the
meaning of words in context,
information in the passage, and
information implied by the passage
but not explicitly stated.
Sentence completion
A sentence is presented with
some words omitted.
The examinee selects words to
complete the sentence.
Mathematics
All questions are multiple-choice or
require a numerical answer.
Numerical operations
Calculation
Calculators are permitted.
Algebraic functions
Algebra problems
Covered to the level of a
second-year high school course.
Geometry
Geometric problems
Questions cover Euclidean
geometry up to three-dimensional
solids
Statistics and probability
Questions in elementary
statistics and probability theory
Questions include interpreting
graphic displays of data.
had expressed, and Chauncey was prepared
to respond.
Chauncey established the Educational
Testing Service (ETS], a nonprofit corpora¬
tion, in 1948. The ETS took over administra¬
tion and development of the SAT, and con¬
tinues to do so sixty years later. The SAT has
become a watchword for students intend¬
ing to enter college, and ETS has become a
major developer of both yearly updates of
the SAT and numerous other educational
and professional tests.19
The present SAT is taken in two parts.
The first part, which the ETS refers to as the
SAT-I, deals with skills that entering college
students are expected to have, but that are
not tied to specific optional parts of the cur¬
riculum. The second part of the SAT con¬
tains subject-matter specific tests. We will
generally be concerned only with the first
part, which will be referred to simply as the
SAT unless there is a reason to distinguish
between parts I and II.
19 Lemann, 1999.
The SAT is divided into three sections.
One, introduced in 2005, is a writing test
in which examinees are given twenty-five
minutes to write an essay on a specified
topic. The inclusion of a writing exami¬
nation is a good example of the demands
placed on a screening test, as opposed to
a test designed to evaluate general mental
competence. Essay writing is a not a uni¬
versally required skill in our society, but it
is a skill that is used in college. The writ¬
ing examination was included in response
to complaints by college and university fac¬
ulty members that some students who had
received acceptable scores on the pre-2005
SAT turned out to have poor writing skills.
As yet there is insufficient data to show
how well the writing test score relates either
to the other tests or to the later performance
of college students.
The other two parts of the SAT
are the “critical reading” (formerly verbal
comprehension] and mathematics sections.
Table 2.6 describes the sorts of questions
asked. Scores on the sections can range from
200 to 800. Mean scores on the individual
THE TESTS
43
Table 2.7. The Subtests of the Armed Services Vocational Aptitude Battery (ASVAB)
Name of Subtest
Commentary
General science
Questions about science roughly at the level of courses taught in
US middle schools.
Arithmetic reasoning
Numerical calculations.
Word knowledge
A vocabulary test.
Paragraph comprehension
Read a paragraph and answer questions about it.
Mathematics knowledge
Apply simple mathematical formulae.
Electronics information
This and the next test are designed to identify enlistees who would
be good candidates for training in appropriate technical specialties.
Auto and shop information
See above.
Mechanical comprehension
Tests the understanding of common mechanical problems.
Object assembly
Evaluates skill in seeing how objects should be assembled from
diagrams.
sections are in the 550 range (varying some¬
what from year to year), with a standard
deviation of approximately 100. The overall
score is determined by adding the two sec¬
tion scores, so 1600 represents a perfect score
on the SAT. The writing test is scored sepa¬
rately.
The questions on the SAT are gener¬
ally harder than similar questions on tests
intended for the general population, such
as the WAIS and comparable group tests,
because the SAT is intended for the top two-
thirds of high school graduates, rather than
for the population at large.
2.3.4. The Armed Services Vocational
Aptitude Battery
The Armed Services Vocational Aptitude
Battery (ASVAB), administered and devel¬
oped by the personnel branch of the US
Department of Defense, is designed to
evaluate potential enlistees and to identify
candidates for training in the military’s many
occupational specialties, ranging from cooks
to computer technicians. Like the SAT, the
ASVAB is updated regularly. Table 2.7 lists
the subtests of the ASVAB as of 2008. The
tests may be administered either in paper-
and-pencil form or, as is now usual in recruit¬
ing, via an interactive computer system.
The Armed Services Qualification Test
(AFQT) score is a weighted combination
of the Arithmetic Reasoning, Mathematical
Knowledge, Paragraph Comprehension, and
Word Knowledge scores of the ASVAB.
This makes the AFQT a combination of
assessments of mathematical and written
verbal skills, as is the SAT total score. There
is a correlation of .82 between overall scores
on the SAT and the AFQT.20 The SAT,
being intended for a more cognitively pow¬
erful audience, is considerably harder than
the AFQT.
The special-topic subtests of the
ASVAB - General Science, Electronics and
Auto and Shop Knowledge, Mechanical
Comprehension, and Object Assembly -
are used as indicators to see if a given recruit
qualifies for specialized training courses
(e.g., electronics repair).
Each service establishes a minimum
AFQT score that must be achieved for
enlistment. The Navy and Air Force, both
of which have a large number of technical
billets to fill, generally set higher minimums
than the Army and the Marines. Recruit¬
ment standards in all the services vary from
time to time, due to the needs of the services
for new recruits, compared to the talent lev¬
els in the applicant pools.
20 Frey and Detterman, 2004.
44
HUMAN INTELLIGENCE
In addition to its use in military recruit¬
ing, the ASVAB has been used by the
Department of Labor in research on the US
workforce.
2.4. Single-Format Tests
While tests like the SAT, CAT-3, and
ASVAB are economical from the viewpoint
of the institutions that use them, they still
require several hours of the examinee’s time.
Industrial personnel screening and psycho¬
logical research programs have a need for
tests that make fewer demands on the peo¬
ple who take them, even though the result¬
ing examination may be less detailed.
2.4.1. Baddeley’s Three-Minute Reasoning
Test as an Illustration of a Speeded Test
One strategy for designing a quick test of
cognitive skill is to shift from measuring a
person’s accuracy in answering reasonably
complex questions, to measuring the speed
with which that person can answer simple
questions. The argument for doing so is that
those people who can answer hard prob¬
lems, given time, are also likely to be able
to answer easy problems, quickly.
The British psychologist Alan Baddeley
developed a test of verbal intelligence based
on a prototypical linguistic act, determin¬
ing whether a sentence correctly describes a
visual situation. Questions on this test take
the form
A B The A is before the B True _ False _
B A The A is not after the B True _ False _
Individually, these questions are easy; no
one who reads English should ever get one
wrong. The sentences vary in linguistic com¬
plexity, although that complexity is well
short of the complexity that might tax a pro¬
ficient reader. While only two pictures are
possible, A B or B A, there are eight possi¬
ble questions, composed by varying whether
the question asks about the position of A
relative to B, or B relative to A, whether
the relation specified is before or after,
and whether the descriptive statement is
expressed positively, as in the first exam¬
ple just given, or negatively, as in the second
example. When taking Baddeley’s test it is
not hard to answer the question, but it may
take a little effort to figure out what the
question is.
The score on Baddeley’s test is the num¬
ber of questions a person can answer in just
three minutes. Simple as it is, this score
has a correlation of .6 with scores on ver¬
bal intelligence tests that take over an hour
to administer.21
What Baddeley showed is that the ability
to do certain simple verbal manipulations,
rapidly, is correlated with the ability to do
such complex things as comprehending a
paragraph. Therefore, the quick test can be
used to identify good verbal comprehenders
without asking them to do much compre¬
hension. This is the sort of finding that might
have infuriated Walter Lippmann, because
the test of comprehension only minimally
involves comprehension. The same finding
would have delighted Prof. Boring, because
the test worked.
2.4.2. The Wonder lie Personnel Test
Baddeley’s test was developed for research
purposes, not for the commercial market.
The principle behind Baddeley’s test, eval¬
uating intelligence by seeing how quickly
people answer simple questions, has been
incorporated into a widely used industrial
test, the Wonderlic Personnel Test (WPT).22
The WPT can be thought of as a hyper-
compressed battery in which examinees are
asked to answer simple verbal questions, do
arithmetic, and solve simple logical prob¬
lems. The different types of questions are
mixed up, so examinees have to switch
rapidly from solving one type of problem to
solving another. The WPT consists of fifty
questions similar to those shown in Table
2.8. The score is the number of questions
that can be answered correctly within twelve
21 Baddeley, 1968.
22 Wonderlic, 1992.
Table 2.8. Sample questions illustrating the Wonderlic Personnel Test, revised edition (WPT-R, 2007 revision)
Question 1 Which of the following is the earliest date?
A) Jan. 16, 1898 B) Feb. 21, 1889 C) Feb. 2, 1898 D] Jan. 7, 1898 E) Jan. 30, 1889
Question 2 LOW is to HIGH as EASY is to _ ?__
J) SUCCESSFUL K) PURE 1] TALL M) INTERESTING N) DIFFICULT
Question 3 A featured product from an Internet retailer generated 27, 80, 99, 113 and 213 orders over a 5-hour period. Which graph
below best represents this trend?
/
BCD
Question 4 What is the next number in the series 29 41 53 65 77 __?_
J) 75 K) 88 L] 89 M) 98
Question 5 One word below is underlined. What is the OPPOSITE of the word?
She gave a complex answer to the question and we all agreed with her.
A) long B] better C) simple D) wrong
E
N) 99
E) kind
Note: The items were provided courtesy of Wonderlic, Inc. The underlined word in example question 5 appeared in blue print in the
original. Underlining has been added here to indicate the content of the question.
46
HUMAN INTELLIGENCE
minutes. Research conducted well after the
Wonderlic Test was introduced, in the 1970s,
has shown that the ability to "change set” by
switching between tasks is an important part
of intelligence.
WPT scores are highly correlated with
the full-scale IQ score of the much longer
WAIS. This makes the WPT an attractive
test for personnel screening, especially in
situations where there is a concern that
the applicant meet a minimum standard
of cognitive ability. In one high-profile
application, the National Football League
has used the WPT to screen players trying
out for a career in professional football.
Having a high WPT score will not get you a
lucrative American football career. Having
a low WPT score may make the scouts and
coaches think twice, for there is a cognitive
component to the life of a professional
player.
2.4.3. Raven's Progressive Matrices Tests:
A N on-speeded, Single-Format Test
We next examine tests that use the Pro¬
gressive Matrix item format, which is a for¬
mat designed to evaluate inductive reason¬
ing. These tests are widely used in research,
and have also been used in personnel selec¬
tion, particularly in Europe.23 Progressive
matrix tests represent something of a con¬
trast between an emphasis on speed and
accuracy. The item format can be used to
present problems that vary widely in diffi¬
culty, so part of the score depends on how
hard a problem the examinee can solve. The
tests are usually given with a time limit, so
the score also depends upon how quickly the
examinee can find a solution.
Progressive matrix tests were devel¬
oped by John C. Raven in 1938. Raven
had studied with Charles Spearman, who
believed that the defining characteristic
of intelligence was the ability to detect
and manipulate complex patterns appear¬
ing in observations.24 He referred to this
23 J. Raven & J. Raven, 2008.
24 Spearman, 1923. Spearman’s ideas are further dis¬
cussed in Chapter 3.
ability as eduction; most modern researchers
would probably say "inductive reasoning."
Raven constructed three tests to evaluate
inductive reasoning: the Colored Progres¬
sive Matrices (RCPM) test, for children, the
Standard Progressive Matrices (RSPM) test,
for general use, and the Advanced Progres¬
sive Matrices (RAPM) test for evaluating
people of above average intelligence, such
as college students. The progressive matrix
format has since been adopted for subtests in
a number of battery-type tests, including the
WAIS-IV.
Figure 2.2 shows two progressive matrix
questions. The examinee is shown a 3 x 3
matrix, in which each entry is a geometric
figure. The entries differ in some systematic
way along both the rows and the columns.
The lower right-hand entry (position (3,3)
in matrix notation) is blank. The task is to
select one of the alternative answers at the
bottom in order to complete the matrix. In
order to do this the examinee has to solve
one problem defined by variation across
rows, then keep the solution “in the head"
while working on another problem defined
by variation across columns. As Figure 2.2
illustrates, problems can vary considerably
in difficulty.
The Raven tests take about forty-five
minutes to complete, making them a cheap
alternative compared to the SAT or the
WAIS. A more recent form administered by
an interactive computer system takes even
less time.
Raven’s own matrix tests always use fig-
ural patterns, as in the example. Progressive
matrix tests can also be constructed using
numerical and verbal material. What num¬
ber should be used to fill in this matrix?
1 2 3
2 4 6
4 8
It has been claimed that progressive
matrix tests are the best measures of gen¬
eral intelligence that we have.25 The claim
is based on statistical evidence indicating
25 Jensen, 1998.
THE TESTS
47
L
O
9
1 00 1
Q—
1
II
9 |
1 1 1 9 9 1
1 I'll PM
X
MM
Mn
X
h©H
-X-
X
MM
X
X
X
X
Figure 2.2. Two progressive matrix items. The
task is to complete the 3x3 matrices by
selecting the appropriate pattern from the eight
patterns below the line. The items have been
made up by the present author, to illustrate the
technique.
that scores on these tests are associated with
whatever common trait underlies perfor¬
mance on all subtests of a test battery, such
as the subtests of the SAT or the WAIS. This
argument will be described in Chapter 4.
A number of researchers have used pro¬
gressive matrix tests in cross-cultural stud¬
ies, and have sometimes made claims about
the relative intelligence of different groups
based upon the results. The argument for
doing so is that the ability to find patterns
in observations is a cognitive ability that
is required in all societies. This is a very
strong claim. It assumes both that the pro¬
gressive matrix item format is appropriate
and that the “Drop in from the Sky” testing
paradigm is an appropriate means of evalua¬
tion outside of industrial and post-industrial
societies. Such claims raise complex issues,
for validity is not an “either/or” issue. A
test may be valid to different degrees in
different societies, and societies themselves
do not fall into precisely defined categories,
such as "developed,” "industrial,” and "post¬
industrial” societies. This matter will be pur¬
sued further in Chapter 11.
We now turn to some general issues that
are raised in the construction and use of the
tests.
2.5. Themes in Testing
There is a great deal of commonality in the
cognitive skills the various tests evaluate.
Here are some of the recurrent themes.
2.5.1. Language Use
Virtually all battery-type tests evaluate lan¬
guage skills. There is a good case for doing
so. Although a person cannot participate
fully in human society without language
skills, there are substantial individual differ¬
ences in the degree to which people pos¬
sess these skills. Everyone learns to produce
and comprehend a passable version of his or
her native language. Very few people reach
the levels of comprehension and expression
illustrated by Shakespeare and Cervantes.
In intelligence testing an important dis¬
tinction is made between language familiar¬
ity and language comprehension. Language
familiarity is usually assessed by a vocabu¬
lary test. Language comprehension is tested
by asking examinees to explain the mean¬
ing of sentences or paragraphs. In theory
one might argue for separate evaluations of
comprehension of the written and spoken
language. The case for doing so rests on
a distinction articulated by the evolution¬
ary psychologist David Geary, who main¬
tains that we have genetically determined
“primary” mental capacities, common to all
societies, and socially acquired “secondary”
capacities that are required in only some of
them.26 Use of the spoken language is clearly
a primary capacity. Normal children learn to
speak without explicit tuition, simply being
reared by speakers of the local language. Lit¬
eracy is a secondary skill. Reading and writ¬
ing are acquired through instruction, and the
26 Geary, 2005.
48
HUMAN INTELLIGENCE
success of this instruction varies greatly. It is
possible, for instance, to be competent in the
spoken language and still be nearly unable
to read [dyslexic). Adults can suffer brain
injuries that render them unable to read or
write even though they can still speak. This
is called acquired dyslexia. This contrasts
with developmental dyslexia , which is mani¬
fested when an apparently healthy child has
abnormal difficulty learning to read.
The dyslexias are an exception to the
general rule that skills in spoken and writ¬
ten language are highly correlated, a find¬
ing that has been obtained in two different
populations that have considerably different
language skills: American college students
and military enlisted personnel.27 Finding
that the same relationship occurs in differ¬
ent populations provides strong evidence for
the generality of the relationship.
2.5.2. Visual-Spatial Reasoning
Almost all battery-type tests that are
avowedly evaluations of intelligence mea¬
sure some form of visual-spatial reasoning.
The tasks used vary in the extent to which
they involve perceptual or reasoning pro¬
cesses. At the perceptual end, some tests
require the identification of a pattern hid¬
den within another or identification of two
patterns as the same, different, or perhaps
one as a mirror reflection of the other. An
example problem is shown in Figure 2.3.
The reasoning end is represented by pro¬
gressive matrix problems and similar tasks,
in which a [possibly abstract) pattern must
be extracted from visual displays.
There has been a good deal of contro¬
versy over whether visual- spatial reasoning
tests should be included in test batteries
used in educational and industrial person¬
nel selection programs, such as the SAT
and the ASVAB. At present [2010) nei¬
ther battery assesses visual-spatial reasoning.
Those who would include such tests argue
that visual-spatial reasoning is an important
27 Palmer et al. (1985) present data for college students;
Sticht (1975) presents data for military enlisted per¬
sonnel.
part of cognition in some courses of study,
such as architecture and engineering, and
in a variety of mechanical trades of use
to the services.28 Visual-spatial reasoning is
assessed in some detail in personnel selec¬
tion for some occupations, especially in
aviation.
The argument against assessing visual-
spatial reasoning is that there are many areas
where the ability is not required. Visual-
spatial reasoning is itself a complex domain.
There are substantial male-female differ¬
ences in performance on some tests of visual-
spatial reasoning. Therefore, if a person¬
nel selection program gave undue weight to
visual-spatial reasoning, the selection system
might be considered biased against women.
2.5.3. Mathematical Reasoning
Many, but not quite all, attempts to evalu¬
ate intelligence include some evaluation of
mathematical reasoning. In the simplest case
this amounts to nothing more than a test of
how rapidly one can do simple arithmetic.
See, for instance, the numerical problems on
the Wonderlic Personnel Test [Table 2.8).
Are tests of mathematical reasoning tests
of one’s academic background, or are they
tests of general intellectual competence?
The answer is “probably a little bit of both.”
Citing Geary and the evolutionary psychol¬
ogists once again,29 humans are genetically
programmed for some aspects of numerical
reasoning, such as the idea of distinguishing
specific numbers of objects rather than mak¬
ing a binary one-many distinction. These
skills are rudimentary mathematics, com¬
pared to the calculus! Acquisition of sub¬
stantial mathematical skill clearly depends
upon having an appropriate cultural back¬
ground. Therefore, in Geary’s classifica¬
tion, mathematical skills beyond a rudimen¬
tary level are secondary skills. Nevertheless,
there is a good reason to assess them. Skill
in mathematics is central to functioning in
developed societies.
28 Hegarty, Just, & Morrison, 1988; Humphreys &
Lubinski, 1996.
29 Geary, 2007a.
THE TESTS
49
□ □On
Figure 2.3. An example of a visual-spatial
reasoning problem. Three of the four figures
numbered 2 through 5 can be made identical to
figure 1 by rotating them in the picture plane
without lifting or turning the figure. Which of
the figures cannot be made identical to figure 1 in
this way?
2.5.4. Deductive and Inductive Reasoning
The ability to reason is widely accepted as
a sign of intelligence. Two different types
of reasoning are recognized, deductive and
inductive reasoning. In deductive reasoning
an examinee is told to assume that certain
statements are true, and is asked to draw
conclusions from them. Examples are
Categorical syllogism: All the girls in Ms.
Jones's class went on the field trip yester¬
day. Ann and Sally are in Ms. Jones's
class. Where were Ann and Sally yester¬
day?
Deduction: John and Sam drink beer only
when they are together. Yesterday John
was in town and Sam was out of town.
Did John drink beer yesterday?
There are logical and empirical argu¬
ments for evaluating abstract deductive rea¬
soning.
Syllogisms and categorical reasoning are
central to Western notions of mathemat¬
ics, law, and rational argument. Mothers tell
children, “If you don’t eat your vegetables,
you can’t have dessert.” Chocolate-loving
children are supposed to draw an appropri¬
ate conclusion.
Empirically, scores on abstract reasoning
tests can be used to predict performance
on other applications of intelligence, such
as paragraph comprehension, mathematics,
and (to a lesser extent) the ability to solve
visual-spatial reasoning problems. Because
abstract reasoning predicts so many other
types of thinking, it is reasonable to believe
that abstract reasoning is either a central part
of intelligence, itself, or that it is closely tied
to something that is.
However, there is a case against stressing
it too much. Although abstract reasoning is
dear to Western academic and scientific cir¬
cles, outside these circles some people think
of abstract reasoning as a sort of word game,
with little intellectual content. Let me give
two examples.
In one of the Sherlock Holmes stories,
"Silver Blaze,” the great detective deduces
that an intruder did not break into a stable
to steal a horse because a watchdog did not
bark in the night. This is an example of the
classic syllogism:
Inductive reasoning is the process of
abstracting general rules from observation of
specific cases. Progressive matrix tests were
developed to evaluate this ability. Inductive
reasoning is also evaluated using other for¬
mats. Here are some widely used ones:
Similarity: Which of the following cities
does not belong in the group: San Fran¬
cisco , Las Vegas, San Diego, St. Paul?
Series completion: Complete the next num¬
ber in the following series: 3:5:7 _ ?
Analogies: Choose the correct answer to
complete the analogy: Black is to white
as right is to: left., error, up, tight.
Premise: A implies B. If there is an intruder
the dog will bark.
Observation: NotB. The dog did not bark.
Conclusion: Therefore, Not A. Therefore,
there was no intruder.
When I used this example in an under¬
graduate class on critical thinking one of my
students said that that was a nice made-up
story, but that watchdogs do not always bark
when strangers appear. Her more general
point was that the real world is far more
complicated than the abstract world of log¬
ical reasoning. She did not see the point in
learning ways of thoughts that were useful
5°
HUMAN INTELLIGENCE
only in the abstract world. She is not alone
in her attitude.
Abstract reasoning is actually proscribed
in some cultures. Consider the example
of the two men who always drank beer
together, given earlier as an example of a
syllogism. The example is a paraphrase of
an item that was used in an anthropological
investigation of reasoning across cultures.30
Rural Liberian tribesmen said that the ques¬
tion was not reasonable, on the grounds that
they did not know the individuals involved,
and one ought not to draw conclusions
about things that one has not experienced
personally.
Both these arguments can also be used
against the inclusion of abstract inductive
reasoning items in intelligence testing. In
addition, there is another objection. The
answer to an inductive reasoning problem
is never uniquely determined. This is shown
by the example in which an examinee was
asked to find the dissimilar city in the set
(San Francisco, San Diego, Las Vegas, St.
Paul}. One could argue that the first three
cities are in the Far West of the United
States, while St. Paul is in the Midwest.
Or that the first three cities have Spanish-
derived names, while St. Paul does not. Or
that Las Vegas is the only city not named
for a Christian saint. There are many ways
in which the four items can be compared;
the question does not specify which one is
to be used.
Similar arguments can be made against
the use of series or analogy problems, includ¬
ing progressive matrix tests. In all cases the
answer is determined by consensus, and the
consensus opinion may be different in dif¬
ferent cultures. Here is another example.
Which two of these animals belong together:
fox, cat , dog?
When University of Michigan students
were asked this question most of them said
that the fox and the dog belong together
because they are both canids. When Central
American Amerindians were asked the same
question their preferred answer was that the
fox and the cat belong together because of
30 Cole et al., 1971.
their similar behavior. The Michigan stu¬
dents preferred a taxonomic grouping; the
Amerindians preferred an ecological one.31
Who is to say which answer is correct?
This example is by no means an iso¬
lated one, restricted to a contrast between
students in a post-industrial society and
members of a (somewhat) traditional agrar¬
ian culture. Richard Nisbett, a professor at
the University of Michigan who has made
extensive studies of cultural influences
on thought, has offered similar examples
involving contrasts between the reasoning
of American and Eastern (Asian) styles
of thought. Nisbett has argued that the
American-European emphasis on focusing
on only the perceived relevant aspects of a
situation, and applying formal logic to those
aspects only, is a marked contrast to an East¬
ern emphasis on being sensitive to the total
context of a problem.52
2.5.5. Another Way to Look at Content:
Aptitude versus Achievement
A distinction is sometimes made between
aptitude and achievement tests. The distinc¬
tion can be illustrated by comparing three
different college admissions tests: the SAT-I,
the SAT-II, and the American College Test
(ACT). The SAT-I is said to stress aptitude,
because its content is not tied to specific
course curricula. The SAT-II and the ACT
contain subtests tied to courses, tests of his¬
tory, literature, science, and mathematics.33
What is the difference?
Achievement is simple enough. You have
successfully studied English, mathematics,
history, physics, or anything else if you know
the relevant facts and principles. But what
is meant by “aptitude?”
31 Lopez et al., 1997.
32 Nisbett, 2009, pp. 162-170.
33 The ACT was originally developed by E. Lindquist,
a psychologist who was involved in the early devel¬
opment of the SAT. Lindquist disagreed vehe¬
mently with Chauncey's decision to emphasize apti¬
tude over ability. He believed that it was more
appropriate to evaluate a person's readiness for fur¬
ther education by determining what the person had
learned up to the point of testing.
THE TESTS
51
According to Richard Snow, an educa¬
tional psychologist from Stanford Univer¬
sity, “aptitude” implies that a person has
a talent for doing something - politics,
music, athletics, and so forth. 54 “Achieve¬
ment” implies that a person already has done
something in a certain field. A person who
has an aptitude for a field should do well
if he or she trains in that field, but having
“aptitude” does not imply that the training
has taken place. You can have an aptitude
for playing music without knowing how to
play the clarinet. But if you have the apti¬
tude, learning to play should be easy.
The SAT-I was once explicitly called a
scholastic aptitude test because it was sup¬
posed to identify students who could be suc¬
cessful if they went to an American college
or university, without stressing the knowl¬
edge acquired in particular courses. By con¬
trast, the SAT-II and ACT test identified
topics, such as mathematics. However, the
SAT-I is not entirely knowledge-free. For
instance, it assumes that examinees know
the English language, but does not assume
that they have had a course in English
literature.
Progressive matrix tests are aptitude tests
that make even fewer demands on knowl¬
edge. However, users of these tests assume
that the examinee understands the basic
testing situation, and there are various
strategies for taking such tests that depend
upon cultural knowledge.
The developer of an achievement test
assumes that all examinees will have had
certain experiences - for example, a course
in American history. The test is intended to
determine what the examinee learned from
the experience. The argument for using
achievement tests as screening devices is
that one of the best predictors of how well
a person will do at learning something in
a new situation is how much the person
has learned in comparable situations. By
this logic, the best predictor of grades in
first-year college physics is a test of how
much the student learned in high school
physics.
34 Snow, 1996.
In educational settings it turns out that
very much the same admission decisions are
made regardless of whether an aptitude or
an achievement test is used. While there
are cases of people who score highly on
the SAT-I and do poorly on the ACT, and
vice versa, at the population level the SAT
and the ACT predict success or failure for
almost the same people.*5 On the whole,
if you have an aptitude for academic stud¬
ies, then you will have learned a lot from
the classes you have already had; if you do
not have the aptitude, you will not. This is
not to deny the fact that late bloomers do
exist; there are people who have the tal¬
ent to succeed in academics but who, for
a variety of reasons, have not learned very
much in high school. Conversely, there are
people who do not do well on abstract apti¬
tude tests, but are quite good at learning
academic material. These cases are excep¬
tions to the rule. Academic achievement
and academic aptitude test scores are highly
correlated.
2.6. Test Creation and Use
We now move from a consideration of the
content of cognitive tests to some more
general issues about how the tests are
developed.
2.6.1. Item Selection and Evaluation
The questions on intelligence tests have
been the butt of quite a few jokes and sar¬
castic remarks. Here is what one popular sci¬
ence writer admitted that he believed prior
to looking into the issue.
the questions on IQ tests are all written by
graduate students from Connecticut and
begin “Teddy leaves Sag Harbor on the
brunchtime jitney ..."
B. Maddox, writing in Discovery,
March 2008, p. 18
35 Koenig, Frey, and Detterman, 2008. These authors
also show that the ACT has a substantial predictive
correlation, .61, with the Raven Progressive Matrices
test, which is in no sense an achievement test!
52
HUMAN INTELLIGENCE
Maddox’ s comment is not too far from
Walter Lippmann’s characterization of the
first Stanford-Binet test, eighty years earlier
(see Chapter 1). Neither criticism addresses
the rigorous procedures used for item selec¬
tion. And, in fairness to Maddox, later in his
article he explains that his initial prejudices
were wrong, and he delves into some of the
issues surrounding intelligence testing in a
reasonable way.
Candidate questions on all intelligence
tests are initially developed either by identi¬
fying a group of people who are thought to
be intelligent to varying degrees, and seeing
what sorts of problems they can solve, or by
generating questions from a theory of what
intelligence is. The candidate questions are
then given to a sample of people who are
thought to be typical of the population for
whom the test is intended. One way to do
this is to insert a new question into an exist¬
ing test. Answers to the new question are not
counted as part of the score on the existing
test, but statistics are gathered on them. If
these statistics meet certain criteria, the new
item is incorporated into the next revision of
the test.
If a test question is a good one, scores on
that question should be positively correlated
with people’s scores on the other items on
the test. Consider an analogy between taking
a test and a high jump competition, where
competitors have to jump over a bar set at
various heights. Suppose a competitor can
jump over (“clear”) a 1.75 m bar (5 ft., 9 in.),
but cannot jump over a bar set to 1.83 m
(6 ft.). We would expect this competitor
to be able to jump over a 1.6 m bar, but
not to be able to jump over a 1.9 m bar.
More generally, if a jumper can clear a bar
at height x, but cannot clear a bar at height
y {y > x), we expect the jumper to clear all
heights lower than x, and not be able to clear
a bar at heights higher than y.
Suppose we wanted to construct a ten-
item test of the ability to solve word prob¬
lems. According to psychometrics , the term
for the science of test construction, we
should look for ten problems that (a) could
be ordered in terms of difficulty, defined by
the percentage of people who can solve each
problem, and (b) behave like the high jump
bar - if a person can solve a problem at dif¬
ficulty level % but cannot solve a problem at
difficulty level y, where y is greater than x,
then the person should solve all problems
with difficulty level less than % and not solve
any problems with difficulty level greater
than y.
What I have just described is an ideal
situation. In practice, there will always be
some people who fail to solve a problem at
one level of difficulty, but do occasionally
solve problems at a higher level of difficulty.
Therefore, a statistical technique called Item
Response Theory has been developed to select
questions that can be thought of as measur¬
ing the same thing A' Items that have this
property are said to be scalable. The point
to remember is that a great deal of care is
taken to select test questions that appear to
be evaluating the same skill, but at different
levels of difficulty.
2.6.2. Differential Impact
Avoiding differential impact is part of item
selection. Differential impact means that an
item is selectively difficult for some sub¬
population of test takers, compared to the
general population. The idea may be best
illustrated by an example, once again using
mathematics word problems.
Baseball problem, version 1. The Detroit
baseball team was ahead of the Seattle
team by one run. But at the very end of
the game the Seattle team scored two runs.
Did the Seattle team or the Detroit
team win the game?
This is a simple arithmetic problem. You
can solve it without knowing very much
about baseball. Now look at version 2.
Baseball problem, version 2. It was the bot¬
tom of the ninth and the Tigers were up
by one. The first Mariner walked, and the
next one up homered.
Did the Seattle team or the Detroit
team win the game?
36 Embretson & Reise, 2000.
THE TESTS
53
For anyone who knows baseball this is the
same simple arithmetic problem as version
1. If you do not know baseball version 2 is
harder than version 1, because version 2 has
been written using baseball jargon. If you do
not know that the names of the Detroit and
Seattle teams are, respectively, the Tigers
and the Mariners, all you can do is guess
what the answer might be. Version 2 has
differential impact on people who are or are
not familiar with baseball.
This example was made up to make a
point. However, it does resemble objections
that have been made in realistic situations.
For instance, on occasion tests have been
attacked for differential impact on African
Americans because they were written using
the vocabulary and syntax of the majority
population.
Differential impact does not mean that
tests and items have to be equally difficult
for all groups. To see why, consider again
the high jump competition. Women, in gen¬
eral, cannot jump as high as men. (As of
April 2010 the men's record was 2.45 m, and
the women's 2.09 m.) If we were to set a
high jump bar at a randomly chosen height
between one and two meters, we would find
that a higher percentage of men than women
could jump over the bar. Nevertheless, high
jump bars do not have differential impact,
because they meet the criterion for equal
scalability in both groups; the difficulty of
jumping over a bar increases with height for
both men and women. The same argument
applies to cognitive tests. Tests and items
do not have differential impact if they scale
equivalently in each group. This means that
the order of difficulty must be the same for
all items, and in addition that the items must
meet some further statistical criteria speci¬
fied by item response theory.
This definition of differential impact is
one based on psychometric theory. Law
courts have accepted definitions of differ¬
ential impact that are somewhat differ¬
ent, and that change from time to time
and from jurisdiction to jurisdiction. The
psychometric definition tells test develop¬
ers what sort of items they should avoid.
Satisfying the psychometric definition will
generally, but not always, also satisfy the
legal definition.
2.6.3. The Distribution of Test Scores
It is often said that intelligence is “nor¬
mally distributed.” The facts are a bit more
complex. We need to distinguish between
test scores, IQ and similar metrics, and the
underlying concept of intelligence as a prop¬
erty of an individual. Again we draw on
an analogy between intelligence testing and
high jumping.
We can think of a high jump competition
in the following way. The jumps are ordered
by height, from the lowest to be considered
to the highest. Competitors try to jump at
each height, and are scored by the number
of jumps they clear. A competitor's score
would not depend on the order in which
jumps were attempted. To see this, consider
a person who has the ability to jump 1.5 m
high, and a contest in which the jumps are,
in order of height, 1 m, 1.5 m, and 2 m. That
person will succeed on two jumps, regard¬
less of the order in which they are presented.
An Olympic-level athlete would almost cer¬
tainly succeed on all three, and if I were to
compete, I think I could manage to clear the
lowest jump.
In a cognitive test each question is associ¬
ated with a level of difficulty, just as the
height of a bar is associated with diffi¬
culty in high jumping. The “raw score” on
a test is the number of questions the per¬
son can answer correctly. (If the test con¬
tains multiple-choice questions there has to
be a correction for guessing, but this is easily
made.)
What would the distribution of scores be
over the population of people who either
compete, in high jumping, or take the test,
in intelligence research? That depends upon
two things; how high we set the bar (or how
difficult we make the items) and how skilled
the competitors (test takers) are.
If we were testing male high school-level
competitors using the 1, 1.5, and 2 m bars,
virtually everyone would clear the lowest
bar, most would clear the middle bar, and
a few would clear the highest bar (a little
54
HUMAN INTELLIGENCE
below the US national high school record).
The distribution of scores would contain
very few o’s, some i's, mostly 2’s, and a
few 3's. We could change the distribution of
scores by changing competitors. The score
for male college jumpers would be mostly
3's, for female jumpers almost all 2’s (for the
highest bar is just under the world women’s
record). Or we could change the distribu¬
tion of scores by changing the height of the
bars. Suppose the bars were set at 1.4, 1.6,
and 1.8 meters. The scores for male college
jumpers would pile up in the 3's, and we
would begin to see more 3’s in the women’s
competition.
We could do exactly the same thing
in cognitive testing, except that we would
manipulate item difficulty instead of the
height of the high jump bar. For a given
population (high school students, military
recruits, etc.) the distribution of raw scores
(numbers of questions answered correctly)
depends upon how many items the test con¬
tains at different levels of difficulty. For a
given test, the distribution of scores depends
upon the distribution of cognitive skills in
the population.
Many cognitive tests are constructed so
that they yield a normal distribution of
test scores in their intended population.
For instance, SAT scores are approximately
normally distributed over the population
of people who apply to American colleges
and universities. This is a selected group
of individuals of higher-than-normal intel¬
ligence. By contrast, scores on the WAIS are
intended to reflect a person’s intelligence
relative to all people in the population. A
student who receives a near-average score
(somewhere around noo for the combined
scales) on the SAT-i would probably have
an IQ score of over 100, simply because the
mean intelligence level of the high school
students who apply to college is higher than
the mean intelligence level of all people in
that age group.
As a result, raw scores on intelligence
tests are (approximately) normally dis¬
tributed because the questions on IQ tests
have been selected to produce a normal
distribution! Contrast this to height, which
(within sexes) is approximately normally
distributed. Since the measurement pro¬
cedure for height is dictated by a theory
of what length is, the fact that height is
normally distributed is a discovery about
nature. The fact that IQ and similar scores
are distributed normally is a consequence of
the way the tests are constructed.
There are marked advantages in requir¬
ing that test scores be normally distributed
over an appropriate population. The statis¬
tical procedures for dealing with normally
distributed scores are well known. If scores
are normally distributed, the standard score
metric (described in Panel 1.1) provides a
convenient way of comparing individuals
to each other in terms of different cogni¬
tive skills, measured on the same popula¬
tion. Suppose that we are told that a par¬
ticular individual had a score of 555 on the
verbal section of the SAT-i and a score of
580 on the mathematics section. This does
not tell us too much, for the scores on the
two parts of the test are not comparable.
By contrast, if we were told that the indi¬
vidual had a standard score of 0.1 on the
verbal test and 0.0 on the mathematics test,
we would know immediately that the per¬
son scored just slightly above the mean on
the verbal test and received the mean score
on the mathematics test.
When scores are normally distributed
there is a well-known translation from stan¬
dard scores to percentiles.57 For instance, for
a normally distributed set of test scores, a
person with a standard score of 1 will have
a score that is above the scores of approx¬
imately 85% of the population. Such inter¬
pretations are often useful. Since IQ scores
are simply a translation of standard scores to
a scoring algorithm with a mean of 100 and
standard deviation of 15, compared to stan¬
dard scores with a mean of o and a standard
deviation of 1, the same sorts of statements
about distributions can be made for scores
using the IQ metric.
37 The same thing would be true for any known dis¬
tribution of test scores, but the conversion may be
more complicated. The normal (Gaussian) distribu¬
tion is, by far, the most widely used.
THE TESTS
55
2.6.4. Scoring: An Alternative
to Using Raw Scores
Raw scores are easy to compute and
understand. Item Response Theory (IRT),
mentioned earlier, provides an alternative
method of scoring that has the advantage of
depending much less on an arbitrary selec¬
tion of item difficulties in order to force
a normal distribution of scores, but has
the disadvantage of being much harder to
understand] I shall present the chief results
of IRT, without attempting to present the
mathematics.38
Return, for the last time, to the anal¬
ogy between intelligence testing and a high
jump competition. This time, though, sup¬
pose that the judges have lost their measur¬
ing devices, so they don't know how high
the bars are] They can still conduct the meet,
and assign sensible scores to the competitors.
Here is how this would work.
We can safely assume that every competi¬
tor has a trait, jumping ability. Every bar,
although of unknown height, has a quality
we will call jumping difficulty. We can mea¬
sure jumping difficulty directly, by seeing
what percentage of people can clear a bar,
even if we do not know how high that bar
is. The insight of IRT is that jumping ability
and jumping difficulty must have the same
scale.
We now make an assumption - that in
some reference population jumping ability is
distributed normally, with a standard score
mean of o and a standard deviation of 1.
Arbitrarily, let us decide that the popula¬
tion of male high school students will be
the reference population. We find the bar
(of unknown height, it doesn't matter] such
that half of all high school students can clear
this bar. By our assumption of the normal
distribution, half of all high school students
have a jumping ability above the mean, and
half below, that is, a standard score of o.
Therefore, the bar that just half the students
38 The basics of IRT were developed in the 1960s
(Birnbaum, 1968). The method did not become fea¬
sible until the advent of modern high-speed com¬
puting. See Embretson & Reise (2000) and E. Hunt
(2007) for further discussions.
can clear must have a jumping difficulty
score of o. By the same token, if approxi¬
mately 16% of the students clear a second,
higher bar, then the second bar must have a
jumping difficulty of 1. Why? Because, by
the properties of the normal distribution,
16% of the population has a standard score
above 1.
If we carry out the above norming proce¬
dure for, say, thirty bars of different heights,
we will have a test of thirty items, each of
which has a jumping ability defined by the
percentage of people in the reference pop¬
ulation who cleared each bar. Note that the
raw scores (number of bars cleared) would
not necessarily follow a normal distribution.
The analogy to cognitive testing is exact.
We take a reference population, assume that
the cognitive ability of interest (intelligence,
verbal ability, etc.) is normally distributed
in the reference population, and infer diffi¬
culty levels for questions by observing the
percentages of people in the referent popu¬
lation who can answer each question.
The resulting test can then be used to
measure ability levels in populations other
than the reference population, using the
scale derived from norming in the reference
population. In the high jumping example,
we could compare the scores of college-level
women competitors to those of high school-
level women competitors, using the scale
established by the high school menl In an
intelligence application, we could compare
the intelligence level of, say, Harvard stu¬
dents to Yale students using a scale estab¬
lished at Stanford.
The only arbitrary assumption is that the
underlying ability is normally distributed
in the reference population. If the refer¬
ence population represents a wide range of
ability, this assumption can be defended.
If intelligence is produced by the cumula¬
tive effects of many different causes - rang¬
ing from inheriting good genes to going to
a good school - and if these causes are
independent of each other and no one of
them has a huge effect, then intelligence will
be distributed normally. This follows from
demonstrations of ways in which the nor¬
mal distribution can be derived.
56
HUMAN INTELLIGENCE
2.6.5. The Importance of Norming
Choosing a reference population is an essen¬
tial step in test construction, regardless of
whether the raw score or IRT procedures
are used to establish test scores. The appro¬
priate reference population depends on the
purpose of the test. The Wechsler tests are
intended to be used on national populations.
Therefore, an attempt is made to obtain a
probability sample of an entire population
(e.g., all Spaniards, Americans, or Germans
for the appropriate version of the test) for
norming purposes. Because these tests are
intended to be applicable to all ages, the
norming sample ought to contain a large
number of people of different ages. This is
an expensive proposition, so at times various
compromises are used. For instance, some
norming studies for the Raven tests were
carried out in just one city, but the city was
chosen because it had values on a variety of
demographic variables that were thought to
be 'Typical” of the country in question.39
The problem of norming is much easier
when a test is intended for a clearly defined
subset of a population. The SAT is intended
primarily for students in the junior or senior
year of an American high school and who
intend to try to enter college, so norming is
carried out in samples of college applicants.
The ASVAB is normed by taking a popu¬
lation sample of all high school students,
because of a legal requirement that the US
military define its enlistment goals based in
part on the distribution of cognitive skills in
the high school population.
When we compare studies using differ¬
ent tests we have to keep in mind the effect
of different tests having been normed in dif¬
ferent populations. Four of the most widely
used tests, the WAIS, the Raven tests, the
SAT, and the ASVAB (or its subsidiary
scale, the AFQT), are not strictly compa¬
rable. Studies have been done in which
the same people take a pair of these tests.
The investigators can then develop proce¬
dures for converting from one test score to
another.40
39 J. Raven, 2000.
40 Frey & Detterman, 2004; Koenig, Frey, & Detter-
man, 2008.
Test developers try to make a test maxi¬
mally sensitive to changes in the underlying
trait (intelligence, verbal ability, or a similar
cognitive trait) in the middle ranges of the
reference population. This means that if a
test is used to study a population that is very
different from the reference population, the
test may not do a good job of discriminating
between people in the new population. To
see this, imagine that for some perverse rea¬
son it was decided that the AFQT and the
SAT should switch roles; colleges and uni¬
versities use the AFQT, while the military
uses the SAT. It would be hard to distinguish
between the “better" and the “best" stu¬
dents applying to universities, because both
groups would be getting very high scores on
the AFQT, which is markedly easier than
the SAT. This is called a ceiling effect. The
military would find it hard to distinguish
between applicants who were “marginal but
acceptable" and ones who were “unaccept¬
able” because both groups would be getting
very low scores on the SAT. This is called a
floor effect.
Of course, no one is going to switch
the AFQT and the SAT. However, changes
in reference populations do occur, and
can have important practical consequences.
Panel 2.1 describes one of them.
2.7. Some Issues Raised by the Use
of Tests in Personnel Selection
How does cognitive testing fit into psychol¬
ogy, as a science, and into the cultural set¬
ting that produced the tests? These ques¬
tions require a broader view than is provided
by a narrow focus on test scores.
2.7.1. Conflicts of Interest Are Inherent
in Personnel Selection
Personnel selection inevitably produces a
tension between the selecting institution
and the applicant. The argument is often
framed as an argument over a test, but ratio¬
nally it should be over the decision process
using the test, not the test itself. The prob¬
lem can be illustraten by the case of an ide¬
alized college admission process.
THE TESTS
57
Panel 2.1. The Changing Definition
of Mental Disability
Intelligence test scores rose steadily for
most of the twentieth century.* This is
true both for raw scores and for the
underlying scores, derived by using the
IRT methodology. The phenomenon is
covered in detail in Chapter 9. For the
moment, just accept the fact. The rise
has had the unexpected consequence of
raising questions about what we mean by
mental retardation.
Prior to 1992 the American Associa¬
tion for Mental Retardation defined retar¬
dation as a child’s having an IQ of
70 or less, measured by the Wechsler
Intelligence Scale for Children (WISC).
New norms are established for the WISC
on a regular basis. The second version,
the WISC-R, was normed on a sample
representative of the US population in
1974; the third version, the WISC-III,
was normed on a similar sample taken
in 1991. But the 1974 and 1991 popula¬
tions were not intellectually the same. It
has been estimated that over that time
period the mean on the full-scale IQ
score for the WISC shifted upward by
about five points. Therefore, the “zero
point" (IQ = 100, standard score = o)
for the two tests was not at the same point
on the underlying trait.
Tomoe Kanaya, a researcher at Cornell
University, and her colleagues showed
that this fact has serious implications for
policies concerning the developmentally
disabled.' Suppose that a child were to
take the WISC-R in 1990. At that point
the WISC-R, normed in 1974, would
be the latest available test. Suppose fur¬
ther that the child obtained a score of 74.
The child would not be classified as dis¬
abled.
Now suppose that the same child had
been examined in 1992, instead of 1990,
using the then-latest version of the test,
the WISC-III. Presumably the child’s
mental capacities would be the same, but
the reference point, the mean intelligence
score, would have been moved up by five
IQ points. Since IQ is determined by the
deviation of a score from the population
mean, the child would receive a score of
69 on the WISC-III, and hence be classi¬
fied as developmentally disabled.
A similar example could be con¬
structed at the other end of the scale.
Suppose that in order to qualify for a
“gifted student" program a child has to
have an IQ of 130. Imagine a child who
received an IQ score of 133 in 1990, using
the WISC-R. The child qualifies as gifted.
If the same child had been tested in 1992,
using the WISC-III, the IQ score would
be 128 and the child would not qualify as
being gifted.
This forces us to think about what we
mean by terms like “mental retardation"
and “gifted.” Do we want to define these
groups by an absolute level of cognitive
power? If so, qualifying scores should be
adjusted so that the same people qualify
as mentally disabled (or gifted), regard¬
less of the population on which the test
is normed. By this definition, if roughly
2.5% of the population of the US qualified
as developmentally disabled in 1974 (IQ <
70 by 1974 standards), then only slightly
less than 1% (IQ < 70 by 1974 standards,
IQ < 65 by 1991 standards) would qualify
in 1991 - not because the cognitive power
of the mentally disabled had changed, but
because the population as a whole had
become more intelligent. The change in
the incidence of diagnosed cases of retar¬
dation would have major public policy
implications for, among other things, the
amount of money needed for programs
for the mentally disabled.
Another argument is possible. Sup¬
pose we define anyone who is in the
contemporary bottom 2.5% of the intelli¬
gence distribution as being mentally dis¬
abled, regardless of where they would
stand in the 1974 population. If we do
(continued)
5$
HUMAN INTELLIGENCE
Panel 2.1 ( continued )
this, then the people who are classified
as retarded using the 1991 test would, in
a sense, be more intelligent than those
classified using the 1974 test, which seems
wrong. However, you could argue that if
the population becomes more intelligent,
everyday living will become more cogni¬
tively challenging. If this is true, then per¬
haps the definition of developmentally
disabled ought to change, because people
who were just able to keep up with soci¬
ety in 1974, those in the IQ 70-75 range
on the 1974 test, cannot keep up with the
more complicated society of the 1990s.
What do you think?
* Flynn, 1984, 2007.
T Tomoe, Scullin, & Ceci, 2003.
College and university applicants present
evidence of their academic accomplish¬
ments and skills, often including their SAT
scores. Figure 2.4 shows the six-year grad¬
uation rates of college students as a func¬
tion of their entering SAT-I scores. There
is a 7:1 difference between graduation rates
in the highest, as compared to the low¬
est, SAT score interval. Admission officers
should always prefer high scorers to low
scorers.
The problem can be stated formally. Let %
be an applicant’s score on an entrance exam¬
ination and let p(x ) be the probability that
an applicant with score x will graduate.
The institution should set a minimum
score, xmi, [m for minimum, i for institu¬
tion) and should reject all applicants with
scores lower than this. The value of xmi
depends upon the benefits to the institution,
Bif if the student graduates, and the costs
Q, to the institution, if the student does not
graduate. For an applicant with evidence x
the expected benefits are p(x)Bif and the
expected costs (1 — p(x))Q. The admission
officer should accept the applicant if
p{x)B, > (1 - p(x))Q, (2.1)
which is equivalent to
pffl > Q
(1 - p{x)) - Bi '
(2.2)
The left-hand side of equation 2.2 is called
the odds on success, and the right-hand side is
the cost '.benefit ratio. The odds on success are
established by the evidence; the cost:benefit
ratio is determined by the costs and benefits,
as seen by the institution. The college should
set Xmi to be the point at which expected
costs equal expected benefits. The institu¬
tion should accept the applicant only if the
score is at or above the institution’s critical
value.
The applicant can reason in a similar
way. There are individual benefits, Ba, for
graduating from a college or university, and
costs, Ca, for attending but not graduat¬
ing. While the applicant and the institu¬
tion should agree on the implications of
the evidence, p[x), and hence agree on the
odds for success, they may disagree (legit¬
imately) about the cost:benefit ratio. The
applicant should establish a critical value,
xc, and should want to enter an institution
only if his or her test score is above this
value. Because the applicant's costs and ben¬
efits are not the same as the institution’s,
the applicant's critical value, xc, may not be
the same as the institution's minimum value,
xmi ■ The applicant and the institution should
agree about acceptance when the applicant’s
scores are high enough to be above both crit¬
ical values, and should agree about rejection
when the scores are below both values. Dis¬
agreements arise if an applicant's scores are
between the two critical values. If the appli¬
cant’s cost:benefit ratio is lower than the
institution’s, there will be cases where the
institution rejects an applicant who, from
his or her perspective, ought to have been
accepted. If the applicant's cost:benefit ratio
is higher than the institution’s, the applicant
THE TESTS
59
599 699 799 899 999 1099 1199 1299
SAT-I score prior to matriculating
Figure 2.4. Six-year graduation rates as a function of entering
SAT-i scores. Source: The College Board, as reported in The
Washington Post, Jan. 27, 2000, p. An. The data does not reflect the
use of a test of writing, which is now included in the SAT-i.
will reject an acceptance offer. In each case
the disagreement is over the costs and bene¬
fits, not the implications of the evidence. In
actual arguments over the use of tests in per¬
sonnel decisions these different arguments
are sometimes confounded.
2.7.2. Missing Evidence: What the Tests
Do Not Evaluate
Arguments about the interpretation of cog¬
nitive tests take three forms. One, which
is an argument largely carried out among
researchers, is over the appropriate analy¬
sis of a particular study. While such argu¬
ments can be important, they are not the
ones that catch the public’s eye. Public con¬
cerns center on two things: systematic omis¬
sion of certain abilities that might be tested
but often are not, and omission of abilities
that may be important to cognition but that
do not fit into the conventional "Drop in
from the Sky” testing format.
ABILITIES THAT COULD BE TESTED BUT
OFTEN ARE NOT
Cognitive tests are designed to cover those
skills that are required for all the situations
where the tests are used, at the expense of
evaluating cognitive skills that are required
in only some of the applications. The SAT,
for instance, evaluates cognitive skills that
are common requirements across the under¬
graduate curriculum, but does not evaluate
skills in, say, music or drawing, which are
required in only some majors. There is also
no opportunity for examinees to demon¬
strate a cognitive skill unless the examiner
has decided that it is appropriate. The extent
to which this is a problem depends upon
the breadth of the applications for which
the test is intended. The problem is reduced
when a test is intended for personnel classi¬
fication in a well-defined setting, such as the
schools or the armed services. In such cases
test developers can, and do, consult with the
institution to determine what skills must be
tested.
This policy, which is sensible in itself, cre¬
ates a problem when a test is either used
for research purposes, to study intelligence
in general, or applied in a personnel selec¬
tion setting outside of the one for which it
was originally intended. In the former case,
insisting that the current format of a test
(or tests) is the definition of intelligence
essentially puts a blinder on research efforts,
because it amounts to agreeing with Boring
6o
HUMAN INTELLIGENCE
that “intelligence is what the test tests/' thus
ruling out conceptual advances. Critics of
intelligence research have claimed that this
sort of thinking has hurt the field.
When a test is used outside the setting in
which it was developed there is a substantial
chance that the test will fail to evaluate some
aspect of cognition that is, indeed, central in
the new application. Robert Sternberg has
referred to this sort of restriction as a failure
to evaluate practical intelligence, knowledge
and problem-solving skills that the examinee
will actually need “on the job.”41
Sternberg’s concern over practical intel¬
ligence is closely related to his, and others',
concerns that cognitive tests typically do not
evaluate social intelligence, that is, knowing
what to do in various social situations. Eval¬
uating social intelligence sounds like a rea¬
sonable goal, but there are problems. A test
can be designed to determine what an exam¬
inee thinks is right to do in a particular social
situation, but that might not be what the
examinee would do. If we are only inter¬
ested in knowledge of the “correct” answer,
then a test in social intelligence is very close
to becoming a test of verbal comprehen¬
sion. Finally, different segments of the same
society may disagree about what the correct
answer is. To take an extreme example, sup¬
pose a question on a social intelligence test
asked what a deeply religious father should
do if his daughter proposed marrying outside
her family's faith. People in different cul¬
tures will have different views about what
the right answer is.
There have been continuing concerns
over the limited scope of tests of visual-
spatial reasoning. Intelligence tests seldom
evaluate examinees’ ability to reason about
moving objects, as might be required in a
task as simple as crossing a street in the face
of traffic, or maintaining one’s orientation
in the environment. There are substantial
individual differences in the possession of
both of these skills.42 While they are hard
to evaluate using the conventional paper-
and-pencil format, appropriate tests can be
41 Sternberg, 2003.
42 Hunt, 2002.
incorporated into computer-controlled cog¬
nitive testing.
ABILITIES THAT DO NOT FIT INTO THE
CONVENTIONAL TESTING FORMAT
In normal life, thinking is distinguished by
two important characteristics: problem solv¬
ing is attempted because the solution has a
value in itself, and finding the solution may
require considerable time for planning and
reflection. For example, US taxpayers are
highly motivated to file their returns cor¬
rectly and on time. Preparing a return takes
anywhere from an hour to several working
days. The task will be considerably simpli¬
fied if the taxpayer has taken the trouble,
in advance, to organize records of income
and expenses. Doing so requires foresight,
thoughtfulness, and personal discipline.
A test taker solves problems solely to be
evaluated, outside of the context of normal
events. Minutes, not days, are allowed for
problem solving. There is no way to evaluate
the examinee’s skill in executing behaviors
that, by their nature, take time. What might
some of these behaviors be?
Learning: Learning, by definition, is
change in behavior over time. Test develop¬
ers may minimize opportunities for learning
of content during a test because they want
the difficulty of a question to be determined
by the characteristics of that question, rather
than by what an examinee has learned in
answering other questions. This is called the
item independence property. It is a corner¬
stone of the mathematical basis for modern
theories of testing and scalability. But item
independence defeats any attempt to evalu¬
ate learning.
Dynamic testing is a modification of
conventional testing in which learning is
evaluated.44 The idea behind dynamic test¬
ing is that the examiner should first deter¬
mine the level of difficulty of the hardest
problem that a person can solve. This is
said to establish a zone of proximal devel¬
opment. The examiner then provides hints
and suggestions that the examinee can use
43 Allahyar & Hunt, 2003; Hegarty & Waller, 2005.
44 Lidz & Elliot, 2000.
THE TESTS
6l
to solve more difficult problems. The proce¬
dure is continued until a point is reached
beyond which the examinee cannot go,
even with the examiner’s support. The level
of achievement reached after coaching is
assumed to be the examinee’s true ability.
Dynamic assessment is closely related to
a form of educational testing called forma¬
tive assessment .45 Formative assessment has
been shown to aid in learning complex con¬
cepts (e.g., introductory physics) in some
circumstances.46
The value of either dynamic or forma¬
tive assessment as an evaluation instrument
is unclear. To establish the validity of this
sort of testing as an assessment it would
be necessary to show that scores obtained
after coaching are better predictors of future
performance than scores obtained prior to
coaching. As far as I know, this has not yet
been done.
The ability to reflect: Conventional testing
does not allow any time for reflection over
hard problems. When the Enlightenment
philosopher Jean-Jacque Rousseau defined
intelligence he placed the ability to stick
with a problem, over time, on a par with the
ability to think quickly. Reflection and per¬
sistence were the characteristics of Abraham
Lincoln’s style of thought.47 Should a testing
paradigm penalize people like Rousseau and
Lincoln?
Creativity: Creativity is close to “ability
to learn” on people’s lists of desirable men¬
tal traits. The commonsense definition of
creativity is that it is a talent for produc¬
ing novel and admirable mental products.
Biographical studies of people whom we all
agree were creative (e.g., Picasso, Mahatma
Gandhi, Einstein) invariably mention the
time that they spent thinking about the
issues before them.48 This cannot be eval¬
uated in the conventional testing paradigm.
There have been attempts to develop
“creativity” tests that do fit into the nor¬
mal testing paradigm, by asking examinees
45 Black and Wiliam, 1998.
46 Hunt & Minstrell, 1996; Minstrell, 2001.
47 Goodwin, 2005.
48 Gardner, 1993b.
to respond to unusual requests within a
brief time. Two example questions are “List
as many uses as possible for a brick” and
“Within x minutes, write an essay with the
title “The Octopus’s Sneakers.” Such tasks
evaluate an ability to produce novelty on
demand. But is this creativity?
SOME REFLECTIONS ON THE CRITICISMS
How serious are these limitations of the
“Drop in from the Sky” paradigm? The
answer depends upon what one wants to
use the test for - identifying people who
have a trait, or understanding the nature of
the trait.
Identifying people who have a trait does
not necessarily require that the trait be eval¬
uated directly. Identification is possible if
there is a statistical association between the
trait(s) evaluated in the testing situation and
the trait of interest. It has repeatedly been
shown that the scores people achieve on
conventional tests of cognitive ability pre¬
dict their ability to learn rapidly in both aca¬
demic and nonacademic settings. The rela¬
tion between SAT scores and graduation
rates (Figure 2.4) is one example of such an
illustration.
The same argument applies to creativ¬
ity. People whom society considers to be
creative seldom have low intelligence test
scores.49 People with very high test scores
are more likely to produce creative prod¬
ucts than people with average scores or
even above-average scores. David Lubin-
ski, Camilla Benbow, and their colleagues
at Vanderbilt University tracked the perfor¬
mance of people who, at age thirteen, took
the SAT and achieved scores that put them
in the top 1/10,000 (the 99.99th percentile) of
test takers, for that age group. By their thir¬
ties members of this group had garnered an
inordinate number of patents, well-received
literary works, and appointments to posi¬
tions where creativity was expected.50
Cognitive tests seldom evaluate reflectiv¬
ity and creativity directly, because of the
inherent limits of the conventional testing
49 Simonton, 1984.
50 Lubinski et al., 2006.
62
HUMAN INTELLIGENCE
paradigm. However, people who are good
reflective or creative thinkers may be iden¬
tified by their scores on cognitive tests. The
extent to which this is true is an issue to be
decided by appropriate research, not by an
attack upon or defense of the characteristics
of the test.
2.7.3. Public Policy Issues
Tests used for personnel selection have pub¬
lic policy implications, for the public has a
legitimate concern about how educational
and industrial hiring decisions are made.
Showing that a test is a statistically accurate
predictor of academic or employment out¬
comes only partly answers these concerns,
as the following example shows.
In 2001 the President of the University
of California, Richard Atkinson, proposed
dropping the use of SAT-I from California’s
admission program.51 He was strongly criti¬
cal of the use of abstract reasoning tests, and
especially the use of analogies.
Atkinson pointed out that applicants
could improve their scores by taking a spe¬
cial coaching course, which provides prac¬
tice in solving artificial, out-of-context ques¬
tions, such as analogical reasoning items.
This concerned him for two reasons. One
was that he thought that the prospective
student’s time would be better spent study¬
ing academic topics. The other was that
the coaching courses are fairly expensive, so
using such tests as part of the admissions
process gives an advantage to students from
wealthier families, who could afford to have
their children coached.52
51 Speech to the University of California Senate.
Downloaded from www.senate.ucsb.edu/meetings/
townhall/sat/abstracts/AtkinsonSpeech.pdf,
10 November 2008.
52 Atkinson constructively proposed using the SAT-
II, which is a slightly better predictor of academic
performance at the University of California than
the SAT-I. SAT-II scores are less highly correlated
with family socioeconomic status (SES) than are
SAT-I scores (Geiser & Studley, 2002]. We have
come full circle. Atkinson's concern that the SAT-
I favored the wealthy replaced Conant’s concern,
seventy years earlier, that a test like the SAT-I was
needed to avoid favoring the wealthy]
Generalizing Atkinson’s concerns, when
a test is used for personnel selection appli¬
cants will be motivated to acquire the skills
the test evaluates. Decision makers who
adopt a test have to consider whether they
want to encourage such behavior. It may
be that only some applicants (here, wealthy
ones!] will have the opportunity to acquire
the skills. Do the decision makers want to
construct a system that gives some appli¬
cants an advantage over others?
There is another objection that Atkinson
could have raised, but did not. Tests that are
used to make personnel decisions have to
both be fair (i.e., accurate] and appear to be
fair. The public has little sympathy for appli¬
cants to universities who are turned down
because they cannot demonstrate prepared¬
ness in English or mathematics. Applicants
get more sympathy if the denial is based on
their inability to answer analogy questions
such as Car is to Bus as Horse is to: Train ,
Airplane, Donkey, Stage Coach. Students do
not have to answer such questions once they
are in college, so why should they have to
solve them in order to get in?
Concerns such as these cannot be
answered by appealing to statistics show¬
ing that test scores predict success. The tests
must be perceived as fair in addition to hav¬
ing statistical validity.
2.8. Summary
There is no agency that certifies a cognitive
examination as an intelligence test. Instead
what we have is a collection of tests of
cognitive abilities that, to varying degrees,
assess different aspects of intelligence. This
chapter has presented some representative
ones.
Although different tests have been devel¬
oped under different theoretical rationales,
there is a surprising commonality of content
across all tests. Methods for evaluating ver¬
bal intelligence, quantitative skills, and figu-
ral reasoning appear over and over again.
Some tests are given individually; others
are given to groups of examinees. Beginning
THE TESTS
63
around 1995 there was a movement toward
presenting questions on computer screens.
While this has advantages in terms of test
administration, simply putting items on a
computer does not change the nature of
the psychological traits being evaluated; a
vocabulary test is a vocabulary test.
Reliance on out-of-context, “Drop in
from the Sky” testing severely limits what
can be evaluated. Within this paradigm it is
not possible to evaluate certain intellectual
traits, such as a capacity for learning, reflec¬
tion, or creativity, that certainly ought to be
considered part of one’s intelligence. This is
a matter of serious concern.
When all is said and done, though, the
tests are imperfect but not negligible pre¬
dictors of success in academic and industrial
situations. Validity coefficients range from
.30 to .85.
Perfection is not a reasonable goal. An
individual’s success is not solely determined
by his or her personal characteristics; situa¬
tional constraints are often important. Nor
can a test possibly evaluate all the personal
traits that might be important in every sit¬
uation. The fact that there are limits on
predictive ability does not imply that tests
should be disregarded. Test scores do reflect
intelligence.
CHAPTER 3
On Theory
Then I would feel sorry for the good Lord.
The theory is correct anyway.
Albert Einstein [in reaction to a
question from a graduate student
about how he would have felt if
the theory of general relativity had
not received experimental
support)1
3.1. Overview
This chapter and the next discuss theories of
intelligence. This chapter describes the role
that theory plays in our efforts to under¬
stand intelligence. The discussion is neces¬
sary because there has sometimes been a lack
of clarity about what theories of intelligence
are and how they are to be used. Hopefully
the discussion here will provide some prin¬
ciples to apply in analyzing specific theories,
as is done in Chapter 4.
The current chapter begins with a brief
discussion of theory in general. It then
1 Cited in Shapiro, 2006.
moves to a discussion of some special prob¬
lems facing theories of intelligence. Finally,
I contrast two different goals for theories in
psychology, and consider why these goals
may produce different theories.
3.2. Scientific and Nonscientific
Theories
Outsiders sometimes see science as a process
of collecting facts. Theories are seen as sec¬
ondary. Sometimes they are even derided.
For example, proponents of intelligent design,
the belief that life on Earth is the result
of the actions of some all-powerful being,
have insisted that schoolchildren be taught
that the theory of evolution is “only a the¬
ory.” This is not meant as a compliment to
Charles Darwin. Less obviously, but perhaps
more damagingly because the practice is
so widespread, science courses below the
college/university level too often emphasize
memorization of facts and formulae, while
deemphasizing the theoretical explanations
that are offered to explain why these facts
are facts, and why the formulae work.
64
ON THEORY
65
Theories are the glue that holds scientific
observations together, for they summarize
the chains of cause and effect that scientists
use to understand how the world works. The
theory of evolution, for instance, is a mar¬
velously lucid explanation of why life on
Earth is so diverse. Models of physical phe¬
nomena can be, and are, used to design com¬
plicated aircraft that fly (and land safely) on
the very first attempt. Theory lets us look
back to explain why something happened
the way it did, and to look forward to predict
what will happen under various assumptions
about conditions in the future.
Theories are not unique to science.
Many statements in philosophy, popular dis¬
course, and religion are theoretical state¬
ments. The biblical explanation for plagues
in Egypt amounts to a theory; the ancient
Egyptians acted in a way that displeased
Jehovah, who reacted with plagues of frogs,
locusts, and disease. Subsequently Jehovah
wished that a certain favored people, the
Israelites, be permitted to leave Egypt. The
waters of the Red Sea rolled back to let
the Israelites pass, and then rushed for¬
ward again, engulfing the Pharaoh’ s army,
in order that the Israelites might be safe.
These are statements of cause and effect, and
statements of cause and effect constitute a
theory.
Scientific theories differ from nonscien-
tific theories in three ways. The first is that
they must refer to variables that are measur¬
able in principle. The variables in a theory
do not need to be measurable at the time
that the theory is presented. The history of
modern physics contains several instances in
which theoreticians postulated the existence
of subatomic particles before the technol¬
ogy to measure them had been developed.
However, no variable can be admitted to a
scientific theory if, in principle, that variable
can never be measured. The theory of intelli¬
gent design is not a scientific theory because
it postulates the existence of a being whose
motives and powers influence our world, but
whose nature cannot be known.
The second distinction between scientific
and nonscientific theories is that scientific
theories must be interpretable by objective
means. Inspired gurus are not permitted,
and personal faith is never to be confused
with evidence. This does not mean that
anyone, without training, should be able to
understand a scientific theory. The necessary
material may be hard to master, but the rules
of mastery have to be open and nonmystical.
Consider the following two cases.
Albert Einstein's theory of relativity was
arguably the most important development
in scientific theory in the twentieth century.
When Einstein presented the theory of rel¬
ativity he had to present a chain of assump¬
tions and deductions that other, appro¬
priately trained, scientists could (and did)
follow. The theory was accepted by physi¬
cists on the basis of their analysis of Ein¬
stein's ideas, not because they attributed any
mystic properties to Einstein himself. Con¬
trast this to the decision of Saul of Tarsus -
to accept Jesus as the Messiah and begin a
new career as St. Paul, the Apostle to the
Gentiles. St. Paul never claimed to have ana¬
lyzed Jesus’s philosophy; he claimed to have
received a revelation from God.
The third distinction between scientific
and nonscientific theories is that scientific
theories must account for data. A scientific
theory is vacuous unless it implies that cer¬
tain observable events will happen - or, in
more formal terms, that certain patterns will
occur in data. There is no requirement that
the observations be easy to make, or even
that they be possible at the time that the the¬
ory is presented. Several of the predictions
of Einstein's theory of relativity have only
recently been evaluated, using technolo¬
gies that were not available in his lifetime.
The requirement for prediction does mean
that the theory must be stated with suffi¬
cient precision that, given the opportunity
to observe, we can tell whether the predic¬
tion was confirmed or denied. Newton’s the¬
ory of motion predicted that large and small
objects will fall at the same speed in a vac¬
uum, regardless of their mass. And they do.
The prophecies of the Delphic Oracle, Nos¬
tradamus, and other seers are notorious for
their ambiguity.
66
HUMAN INTELLIGENCE
3.3. Choosing between Competing
Theories
Theories imply observations, not the other
way around. When a theory correctly pre¬
dicts an observation, this is evidence in sup¬
port of the theory, but evidence can never
prove that a theory is correct. However, if an
observation is made that contradicts a pre¬
diction, then the theory has been discon-
firmed. Accordingly, in the abstract, we do
not ever determine the correct theory; all we
do is to eliminate wrong ones, and continue
to believe, with reservations, in those theo¬
ries that have not yet been disconfirmed.
At this point statistics rears its ugly head.
If theories always made exact predictions,
and if we could be absolutely certain that
our observations were accurate, then science
would proceed in the way just described. In
practice, that is not what happens. Predic¬
tions are often made in general terms, and
measurements are never exact. Therefore,
empirical results seldom completely rule out
a theory. Instead, what we find is that the
facts are more compatible with one the¬
ory than with another. This will modify our
belief in the “winning” and “losing” theories,
but not necessarily cause us to accept the
winner and reject the loser. Indeed, it often
happens that as more and more evidence
accumulates our belief in all theories under
consideration falls, and we have to develop
new theories.
One of the most hotly debated topics
in intelligence, the Nature -Nurture debate,
provides a good example. In its pure form,
the “Nature” side of this debate asserts
that intelligence is inherited genetically.
The “Nurture” side states that intelligence
is obtained through experience, with the
corollary that early childhood experiences
are particularly important. Let us look at the
logic of the argument, saving the details for
Chapters 8 and 9.
Galton, who was firmly on the Nature
side of the debate, observed that the emi¬
nent men and women of his own gener¬
ation were highly likely to have had emi¬
nent parents or grandparents. Eventually
he identified families who produced people
of eminence, generation after generation.2
Galton concluded that intelligence is largely
inherited.
The hypothesis that intelligence is inher¬
ited implies that there will be a statistical
association between the eminence of fam¬
ilies from one generation to the next. The
“null hypothesis” is that familial eminence
varies randomly from one generation to the
next. Galton found that the generation-to-
generation associations of eminence were
too strong to be accounted for by ran¬
dom variation. Accordingly, Galton had
a strengthened belief in the hypothesis
that intelligence (and hence eminence) is
inherited.
Galton was correct that the data should
strengthen his belief in the Nature (heredi-
tarian) hypothesis relative to the null hypoth¬
esis of no familial association. The position
relative to the Nurture (environmental)
hypothesis is different. Most people inherit
both their genetic composition and their
social environment from their parents.
Therefore, both the Nature and Nurture
hypotheses predict associations of familial
eminence from generation to generation.
The association that Galton found was evi¬
dence for the Nature or the Nurture hypoth¬
esis against the null hypothesis that fami¬
lies don't matter in achieving eminence, but
it was not evidence that could be used to
decide against either Nature or Nurture.
As a side point, this example shows
that, in general, simply testing a hypothe¬
sis against the null hypothesis is not likely to
be very informative.
Proponents of the Nurture hypothesis
can point to studies that show that children
raised in unfavorable environments perform
poorly in school and on intelligence tests,
at least during the early school years. On
its face, this evidence seems to support
the importance of the environment. How¬
ever, proponents of the Nature hypothe¬
sis can point out, correctly, that such chil¬
dren are not a random sample of children in
their generation. Indeed, there is reason to
believe that many of them are the children
2 Galton, 1869.
ON THEORY
6?
Figure 3.1. Genetic and environmental influences determine the
extent to which the intelligences of any two individuals, A and B,
resemble each other. This, in turn, will determine the resemblance
between test scores and other cognitive behaviors of the two people.
We can only observe correlations between manifest variables - test
scores and other cognitive behaviors. In addition, genetic theory can
be used to specify the degree of resemblance between the genetic
potentials of A and B. The other resemblances between latent
variables must be inferred.
of parents who were not terribly intelligent.
The data does not discriminate between the
hypotheses.
By the middle of the twentieth century
the debate had become more refined. Scien¬
tists were willing to acknowledge that both
heredity and environment do have a role,
and that the issue is not to look for a single
cause for intelligence, but rather to deter¬
mine the relative influences of environment
and heredity. At that point a much more
sophisticated model of heredity began to
be used. It is diagrammed in Figure 3.1. Its
basic elements are (a) that there are both
genetic and environmental components to
intelligence and (b] that for any two indi¬
viduals, the extent to which they share the
genetic component of intelligence will be
determined by their genetic relation, and
the extent to which they share environmen¬
tal components will be determined by their
environmental relation.
While it is not immediately obvious, dia¬
grams such as Figure 3.1 (and their associ¬
ated mathematics] can be used to generate
expected correlations between test scores.
The values of the correlations depend upon
the strength of the connecting links. Some¬
times these strengths can be predicted from
genetic theory. In a study of identical twins,
for instance, the genetic link would have
a strength of 1, indicating identical genetic
constitutions. The link for fraternal twins
would have a strength of .5, because frater¬
nal (dizygotic, DZ] twins share half their
genetic inheritance.
More generally, the contrast between
theories, such as Nature versus Nurture, has
been replaced in modern studies by a con¬
trast between models, which can be thought
of as assumptions about the strengths of par¬
ticular links. The extreme Nature (hered¬
ity] model can be thought of as a spe¬
cific case of Figure 3.1 in which the links
from Environment to Intelligence are absent
(strength = o]. We could also investigate
intermediate models - for example, a model
in which the Environment-Intelligence link
is required to have half the strength of the
Genetics-Intelligence link. In all cases we
68
HUMAN INTELLIGENCE
would come back to a basic test: which of
the many possible models best reproduces
the scores between observed variables, that
is, the link between the four rectangular
boxes (observable variables) in Figure 3.1?
We return to this example in Chapter 8.
For the moment, the important thing to real¬
ize is that the field has advanced from evalu¬
ating broad assertions about the importance
of heredity or environment, to a far more
sophisticated competition between mod¬
els whose parameters indicate the relative
contributions of genetic and environmental
causes of intelligence.
3.4. System Thinking Complicates
the Issue
In Chapter 1, I argued that intelligence is not
a directly observable trait; it is a latent trait
that is inferred from observation of various
manifest behaviors influenced by the trait.
These include scores on IQ tests, school
grades, and records of performance on the
job. No one of these is entirely determined
by intelligence, but each is influenced by
it. This means that the definition of intel¬
ligence is, in a sense, something we make
up, rather than being a directly observ¬
able trait. I also pointed out, in Chapter 1,
that intelligence is embedded within a com¬
plex of other latent variables, including such
things as genetic inheritance, socioeconomic
status (SES), and educational opportunity.
These variables have links between them -
for instance, the link between the socioeco¬
nomic status of one's neighborhood and the
quality of the schools.
When variables are linked into compli¬
cated systems there are both direct and indi¬
rect effects. (These are sometimes called
proximal and distal influences.) Untangling
them can be a problem. Here is an example
dealing with verbal intelligence. Once again,
I simply assert the facts, leaving the details
for a subsequent chapter.3
3 See Bouchard, 2009, for further analysis of this
point.
Reading is one of the most important
skills children acquire in the early school
years. Subsequently, reading skills are used
to facilitate learning. The sooner and more
completely the learning to read-reading to
learn shift is made, the more the child can
get out of the educational system. Learn¬
ing to read is associated with a child's fam¬
ily background. On the average, children
from families in the middle and upper SES
strata arrive at school better prepared to
learn to read than children from a low SES
background. This advantage is continued
throughout the school years. Why?
There is a genetic explanation. To the
extent that SES is linked to genetic poten¬
tial, the children from middle-high SES fam¬
ilies may be better learners simply because
they are biologically smarter. There is also
an environmental explanation. The parents
in middle-high SES families, on the average,
spend much more time reading to their pre¬
school children than do parents in low SES
families. Indeed, some of this reading behav¬
ior comes very close to teaching reading, as
in reading books that tell children that “A is
for apple” and so forth.
We are dealing with a set of skills, embed¬
ded in a system. Furthermore, it is a system
with feedback loops, for there is little doubt
that as children learn to read their parents
respond, and the manner of the response
may vary with SES. The situation is depicted
in Figure 3.2, which is only one of several
possible models. It is forbiddingly complex.
Ultimately, every behavior has to take
its action via the child's genetic poten¬
tial. Parental genetic effects are of two
sorts. Directly, parental genetics determine
a child’s genetics, and the child’s genetic
potential, interacting with the social envi¬
ronment, determines reading skills. Indi¬
rectly, parental genetics will partly produce
parental child-rearing behaviors. These
behaviors, influenced by the child's genetic
potential, will produce the child’s behaviors,
which may include reading or pre-reading
skills. The child’s behavior, in turn, influ¬
ences future parental behaviors, and the
cycle continues. And this is only part of
ON THEORY
69
Figure 3.2. Influences on reading skills. Parental genetics will exert a
direct effect on a child’s genetic potential, and also may influence
the parent’s child-rearing practices (e.g., by increased sensitivity to
a child’s mood). Parental child-rearing practices establish a social
environment that, acting through a child’s genetic potential,
produces a variety of behaviors, including reading skills. These
behaviors, in turn, alter the parents’ behaviors toward the child.
the system! I have not shown the effects
of the parental social environment, which
also influence parental behaviors, and many
other influences besides. All the variables are
enmeshed with each other.
This is the point at which many peo¬
ple begin calling for “controlled experi¬
ments.” To continue the example of reading
skills, it has been shown that extensive pre¬
school programs, ones that are far more
intense (and expensive) than the typical
“Head Start” program, will improve chil¬
dren’s cognitive skills during the early school
years. There is even some evidence that
the improvements last, in much weakened
form, into adulthood. The studies in ques¬
tion involved comparisons between the per¬
formances of children who had participated
in an intensive program compared to a
matched group who had not, the classic
experimental group versus control group
study. The details will be given in Chapter 9.
Such studies show that direct manipulation
of a proximal variable, the child's social
environment, can have an influence on the
development of intelligence. However, this
sort of study does not show that genetic
inheritance can be ignored. The reason that
it does not is that random assignment of
individuals to experimental and control
groups, without measurement of, say,
parental intelligence, effectively destroys
any possible correlation between genetic
potential and the child’s performance, as an
artifact of the experimental design.
Suppose that we improve upon the
intervention design, by measuring parental
intelligence and, by appropriate statistical
manipulations, estimating how much the
observed improvement in the child’s per¬
formance can be attributed to the parent's
intelligence, how much can be attributed
to the intervention, and how much is due
to some interaction between the two. Such
designs have been approached, especially in
the study of adopted children (Chapter 8).
The resulting information is useful. How¬
ever, it does not tell us what the controlling
variables are in the world as it is, for we
would be studying a “world as it might be,”
in which there are massively financed pre¬
school programs.
The proper design for a study of intel¬
ligence depends upon what question we
7°
HUMAN INTELLIGENCE
are trying to answer. It turns out that the
proper theoretical model may also depend
upon the question at hand. Intelligence, as
a concept, can be introduced into different
systems. Some questions deal with control
issues in social systems - either how intelli¬
gence might be controlled within social sys¬
tems, or how changing intelligence might
alter those systems. Studies of the role of
intelligence in education, personnel selec¬
tion, and industrial operations focus on
control issues. Other studies deal with reduc¬
tion issues in biological systems - how intel¬
ligence can be derived from some more
basic phenomena. Studies of the relation
between intelligence and the brain, genetic
influences on intelligence, and studies of the
relation between physiological states and
intelligence are reductionist. We want to
conceptualize intelligence in a way that fits
into the system being studied.
3.5. Intelligence as a Construct
in Social Systems
We need theories and models so that we
can think about how things work. We use
them all the time to understand (not always
correctly] how things like electrical circuits
and mechanical pulley-and-lever systems
function.4 On a grander scale, economists
argue (not always correctly] about the right
model for the world economy, while epi¬
demiologists use models (hopefully cor¬
rectly) to forecast the seriousness of an out¬
break of influenza.
Why should models be so ubiquitous?
David Geary5 has argued that humans have
evolved a capacity to construct models of
the world in order to satisfy an inherited
drive to exert control over their environ¬
ment. In order to accomplish this goal the
various components of the model have to fit
together in a way that we can think about.
This means that we may have to conceptu¬
4 Centner & Gentner, 1983; Hegarty, Just, 8t Morri¬
son, 1988.
5 Geary, 2005, 2007a.
alize the same thing in somewhat different
ways, depending upon the world in which
the thing is embedded. That is certainly true
of intelligence.
Here is a case study.
During the 1970s two Stanford Univer¬
sity educational psychologists, Lee Cron-
bach and Richard Snow, studied aptitude x
treatment interactions in education/ They
concluded that classroom environments that
encourage active exploration and experi¬
mentation benefit students with high cog¬
nitive abilities, while students with lower
cognitive abilities may do better in teacher-
structured learning environments. Their
findings have been replicated in situations
as different as college students learning
statistics7 and elementary school children
learning to read.8
Conceptually, Cronbach and Snow
developed a model of the relationship
between psychometrically defined general
intelligence and educational practices. The
variables they studied, psychometric scores
and teaching methods, were measures and
actions available to teachers. Statements
about, say, the behavioral implications of
the level of neural activation of children’s
forebrains during functional magnetic reso¬
nance imaging (fMRI) would be of little use
in the classroom.
If we were to embed the cognitive behav¬
iors that we call “intelligence" into another
system, we might want quite a different
conceptualization. Imagine a (hypothetical)
study in which the investigator is interested
in the use of drugs to ameliorate cognitive
deterioration during aging. In that situation
relating intelligence to the level of neural
activation in a patient’s forebrain might be
a reasonable thing to do.
If a scientific model is to be used to inform
decision makers — that is, to control a sys¬
tem - then the theory must be stated in
terms of variables that the decision maker
can measure and control.
6 Cronbach and Snow, 1977; Snow, 1982.
7 Shute et al., 1996.
8 Freebody & Tirre, 1985.
ON THEORY
71
Lifetime experiences
Genetic inheritance
Figure 3.3. The levels at which theories of intelligence may be
stated. Rectangles indicate physically measurable variables;
ellipsoids indicate hypothetical variables.
3.6. Reductionism
Using scientific models to control variables
is something of an engineer’s view of the sci¬
entific enterprise. It is valid, it is important,
and it is not the only use of theory in science.
The other use is pure understanding.
Again the system concept is useful. A system
is understood when it is completely closed,
so that its behavior can be understood
entirely from the actions of its elements.
This produces an ordering of scientific top¬
ics: the properties of atoms are derived
from the properties of subatomic particles
and forces; the properties of chemical com¬
pounds are derived from their atomic struc¬
ture; the properties of biological systems
are derived from the chemical and physi¬
cal properties of their components, and so
on. The sequence is called reductionism.
Reductionist analyses have important
implications for the study of intelligence.
David Wechsler, the developer of the
Wechsler tests, pointed out that intelligence
is revealed by behavior.9 Behaviors that
we see as exercises of intelligence (or lack
thereof] must be under the control of the
brain. What is the alternative, the influence
9 Wechsler, 1975.
of good and bad angels? Therefore, theories
of intelligence that relate individual differ¬
ences in cognitive power to individual differ¬
ences in brain action are important as steps
in reductionism. It is not clear that such the¬
ories would be manageable, let alone useful,
in understanding how individual differences
in intelligence influence academic learning
or on-the-job performance. Models of these
phenomena might well take as their primi¬
tive terms concepts such as general intelli¬
gence, verbal comprehension skill, and spa¬
tial orientation ability without any concern
about how these abilities are produced by
the brain - just as Newtonian mechanics
accepts gravitational force as a primitive fact
without further explanation.
Figure 3.3 diagrams the reductionist view
of research on intelligence. All performance
depends on the brain. At any moment the
brain is the product of both a person’s
genetic inheritance and his or her environ¬
ment. "Environment” has to be interpreted
in the broadest manner. It does not just refer
to obvious physical factors, such as nutri¬
tion or injury. It also refers to social factors,
including education. Why? Because learn¬
ing produces physical changes in the brain.
This includes the storage of information,
but there is more to it than that. The same
72
HUMAN INTELLIGENCE
external task may be processed by the brain
in different ways before and after learning.
While the brain is ultimately the only
thing that actually does something, it is often
useful to deal with the results of brain action
at the functional level rather than at the
level of brain action itself. To illustrate, con¬
sider how a person recognizes his or her
country's flag. In principle, it would be pos¬
sible to describe flag recognition in terms
of neural action in the occipital, temporal,
and frontal lobes, but in practice it is sim¬
pler to say that visual pattern-detection and
memory-recognition processes are involved.
Individual differences in performance on the
complex task, flag recognition, can then be
related to individual differences in vision
and memory. Failure to recognize the flag
could be due to a distortion of either visual-
recognition or memory-retrieval processes.
Both processes are theoretical entities; there
is a sense in which they do not exist. That
is why information-processing functions are
shown in ellipses in Figure 3.3, while phys¬
ically observable brain and behavioral pro¬
cesses are shown in rectangles. Nevertheless,
there is a place for descriptions of theoretical
processes.
The top of the figure specifies two types
of physically observable behaviors - per¬
formance on psychological tests and per¬
formance in life activities. In the physi¬
cal sense both of these are produced by
the brain. Conceptually, though, we can
think of both behavior in testing situa¬
tions and behavior everywhere else as being
produced by information-processing capa¬
bilities combined with previously acquired
knowledge.
The double-headed arrow between
behaviors on tests and behaviors elsewhere
indicates that the two different types of
behavior are correlated, but that there is
no causal link between them. Psychologi¬
cal test scores do not, in themselves, cause
anything. (The social interpretation of a test
score may be causal, but that is another
issue.) What test scores can do is to indicate
possession of knowledge and information¬
processing capacities that control behavior
in life outside of the testing session.
The next three sections illustrate these
principles by giving brief descriptions of
studies that deal with issues at each of the
three levels of explanation.
3.7. A Study at the Psychometric Level
Psychometric theories deal with the dimen¬
sions of individual ability that are believed
to underlie performance on intelligence and
personality tests. The goal is to simplify the
tremendous amount of data contained in
batteries of tests of cognition or personality,
such as the various battery tests described
in Chapter 2, by showing that the data from
test batteries can be understood in terms of
individual variations along a small number of
dimensions of mental ability, called factors.
Some details of the technique will be
explained in Chapter 4. Here we concen¬
trate on the results, using, as an example,
a model that will be featured prominently
throughout this book.
Two University of Minnesota researchers,
Wendy Johnson and Thomas Bouchard, had
available data from a study in which 436
adult participants had taken 42 different
tests.10 There were over 16,000 numbers in
the data set, obviously more than a per¬
son could think about coherently. However,
Johnson and Bouchard found that a large
part of the individual variation on the forty-
two tests could be captured by individual
variation on just four factors: general intelli¬
gence, verbal skill, perceptual skill, and the
ability to manipulate images “in the mind's
eye,” that is, to imagine how something
seen from one perspective might look from
a different perspective. They referred to
their model as a general intelligence- verbal-
perceptual-rotation (g-VPR) model.
Johnson and Bouchard's work is typical
of hundreds, if not thousands, of psychome¬
tric investigations showing that variation on
many tests can be summarized by variation
along a small number of basic dimensions.
Psychometric studies also provide measure¬
ments that are useful in understanding and
10 Johnson & Bouchard, 2005a.
ON THEORY
73
controlling social systems, most notably
in personnel selection. The SAT, AFQT,
and WPC tests described in Chapter 2 are
derived from psychometric studies, as are
almost all college entrance and industrial
screening tests.
Psychometric models are less useful when
the task is to explain individual differences
on specific cognitive tasks. To illustrate why,
we take a look at an act of verbal compre¬
hension.
Here is an excerpt from a newspaper
editorial:
The citizens of Washington and the nation
are spending millions and millions of dol¬
lars on efforts to protect salmon and save
them and the ecosystem they support from
the threat of extinction.
These goals are being hurt by California
sea lions that have outsmarted the system
and now wait patiently for the fish to be cor¬
ralled into the ladders at Columbia River
dams.
Bellingham (Washington) Herald,
April 11, 200J
From the psychometric perspective, in order
to understand the editorial you have to have
a certain level of verbal ability - in order to
understand the passage at all - and a certain
level of general reasoning ability to realize
that the problem is that sea lions are eating
too many fish to sustain the fisheries. This
fact is not stated directly. It has to be inferred
from an analysis of the text plus some gen¬
eral knowledge about sea lions, predators,
and prey.
Psychometric research could be used to
calculate the probability that a person with
verbal ability level x and general reasoning
ability level y would have probability z of
understanding the passage. Something like
this sort of analysis is used to design instruc¬
tion manuals, where the prose has to match
the reading and reasoning skills of the peo¬
ple who will read the manual. Psychometric
analyses do not tell you what readers have
to do in order to comprehend a manual. For
that sort of analysis we must move to the
information-processing level.
3.8. A Study at the
Information-Processing Level
An examination of what has to be done to
understand the editorial reveals that, with¬
out effort, we do quite a bit of informa¬
tion processing. Retrieving word meanings
is only part of the task. It is complicated
enough, for both semantic and syntactical
information must be retrieved, and ambi¬
guities must be resolved. Does the word
Washington in the editorial refer to the city,
the state, or the president? Then even heav¬
ier information-processing demands appear
during the analysis of sentences and of the
text as a whole.
The comprehension process has to oper¬
ate on words as they appear. However, the
meaning of a word, in context, often can¬
not be determined until following words are
received. The initial the in the editorial has
no meaning until it is attached to citizens. The
citizens of Washington and the nation specifies
which citizens and resolves the ambiguity
of Washington. Next comes are. It appears
that a descriptor is going to follow, that the
collective just described (The citizens . . . ] is
going to be equated with something, as in
The citizens of Washington and the nation
are taxpayers. This interpretation is dashed
when spending occurs, for now are must be
interpreted as an auxiliary verb. Are spending
is not a complete statement; the comprehen-
der must find out what is being spent. The
words millions and millions cannot be inter¬
preted until dollars is encountered.
As this analysis shows, a language com-
prehender must have a cache for holding
information to be integrated with informa¬
tion not yet received. In the current jargon
of psychology, the cache is called working
memory. It is an example of an information¬
processing concept, in the sense that it refers
to an abstract ability to manipulate pieces of
information in the mind, without specifying
what these pieces of information mean in
the world outside the mind.
The result of the construction is a text
model summarizing what the text says
explicitly. The text model alone is not
enough! In order to reach full understanding
74
HUMAN INTELLIGENCE
the comprehender must construct a situation
model of what the text means in the context
in which it is presented.11 This means going
beyond the text to understanding things that
are only referred to obliquely. The editorial
is about sea lions eating salmon, but that is
not stated explicitly]
Clearly working memory is central to the
comprehension process. It follows that indi¬
vidual differences in the capacity of working
memory ought to be related to individual
differences in verbal comprehension. And
they are. Marcel Just and Patricia Carpenter,
professors at Carnegie-Mellon University,
together with their colleagues, have con¬
ducted a substantial series of experiments
investigating the link.12
In order to measure working memory Just
and Carpenter used a procedure called the
memory span task d5 In a memory span task
an examinee is asked to read a fairly compli¬
cated sentence aloud, and then remember
the last word. Another, unrelated sentence is
then presented. After k sentences the exam¬
inee is asked to recall the words. Just and
Carpenter offer the following example, with
k = 2:
Read:
When at last his eyes opened , there was
no gleam of triumph, no shade of anger.
The taxi turned up Michigan Avenue
where they had a clear view of the lake.
Recall:
The examinee should then recall the last
two words: anger, lake.
A person’s memory span measure is
defined as the highest number of sentences
he/she can read and still recall all the ending
words. The argument is that working mem¬
ory is taxed by having to hold the ending
words of previous sentences while the cur¬
rent sentence is processed. Thus the mem¬
ory span is a measure of the capacity of
11 Kintsch, 1998.
12 Just & Carpenter, 1992; Just, Carpenter, & Keller,
1996.
13 Daneman & Carpenter, 1980, 1983; Daneman &
Merikle, 1996.
working memory. College students’ mem¬
ory spans vary from 2 to 5.5, averaged over
participants.14
The Carnegie group found that memory
spans are associated with the ability to com¬
prehend text.15 One of the comprehension
examples they used was the ability to detect
ambiguities, as in
The experienced soldiers warned about the
dangers. . . .
which could mean either that the soldiers
warned some to-be-specified people or that
the soldiers themselves had been warned by
some unspecified person, as in
The experienced soldiers warned about the
dangers conducted the midnight raid.
This contrasts with a phrase like
The experienced soldiers spoke about the
dangers. . . .
where the soldiers are unambiguously doing
the speaking.
Just and Carpenter argued that only peo¬
ple with high memory span could afford
to keep two interpretations (alternative text
models) in mind, and that this allowed them
to achieve a better understanding of com¬
plicated texts than could be achieved by
people with low memory span, and hence
low working-memory capacities. They then
conducted a series of experiments showing
that their conclusion was correct. For exam¬
ple, high-span people read sentences faster
than low-span people. However, high-span
people read sentences containing ambigu¬
ous verbs, like warned in the illustration,
more slowly than they read sentences with
unambiguous verbs. Low-span people read
both types of sentence at about the same
rate. Evidently only the high-span individu¬
als noticed the ambiguity and carried both
meanings forward until the ambiguity was
resolved.
This work is clearly reductionist. A psy¬
chometric construct, individual differences
in verbal comprehension, was related to an
14 Just and Carpenter, 1992.
15 See, especially, MacDonald, Carpenter, & Just, 1992.
ON THEORY
75
information-processing concept, working-
memory capacity.
3.9. Studies at the Brain Level
Explanations of behavior that use infor¬
mation-processing concepts such as
working-memory capacity or retrieval
of word meaning are less abstract than
explanations based on terms like verbal
ability or general reasoning ability, but
they are still abstract. There has always
been considerable interest in explaining
individual differences in behavior in terms
of individual differences in brain structures
and processes. Historically, this was done
by drawing inferences about normal behav¬
ior by analogy to extreme differences in
behavior associated with damage to the
brain, as in J. P. Das’s development of
tests for normal children based on Luria’s
observations of the results of brain injury.
Today’s advanced medical technology has
made it possible to extend the approach,
because we can take a limited look at
the brain in vivo , as people are thinking.
Chapter 7 reviews the current state of
knowledge in this rapidly evolving field.
Here a few quick remarks are in order,
to show how models based on discoveries
in the brain sciences fit into theories of
intelligence.
The new technologies fall into three
broad classes. A variety of imaging tech¬
niques have been developed that permit
direct looks at structures in the living brain,
at a much finer level of detail than provided
by the X-ray imaging technique developed
early in the twentieth century. (The X-ray
method itself has become far more accu¬
rate.) Other technologies, and especially a
technique called functional magnetic reso¬
nance imaging (fMRI), provide information
on metabolism, and hence energy use, in
different regions of the brain. These tech¬
niques have good spatial resolution, but,
because metabolic processes themselves are
slow, and hence reflect the brain’s use of
energy over a period of seconds, temporal
resolution is poor. Techniques for recording
the electrical traces of neural activity, such
as the electroencephalogram (EEG), were
developed in the mid twentieth century,
but today’s capability for recording electrical
events is greatly enhanced over what it was
as recently as the 1990s. Electrical recording
techniques can detect events at the millisec¬
ond level but have relatively poor spatial res¬
olution.
To give an idea of the current state of
affairs, here are some results related to lan¬
guage processing.
Just and Carpenter’s Carnegie-Mellon
group16 participants read sentences of vary¬
ing syntactic complexity, such as
The reporter attacked the senator and
admitted the error.
The reporter that attacked the senator
admitted the error.
The reporter that the senator attacked
admitted the error.
while fMRI images were being taken of their
brains. Metabolic activity increased in two
areas previously known to be involved in
language comprehension - a region in the
left frontal cortex known as Broca’s area,
and a nearby region in the left temporal
region known as Wernicke’s area. As sen¬
tences increased in syntactic complexity the
number of areas of the brain showing acti¬
vation increased. Further studies of this sort,
using linguistic and nonlinguistic stimuli,
have shown that as the difficulty of a task
increases, metabolism increases in the areas
of the brain required to solve the task. How¬
ever, studies of individual differences have
shown that high performers (e.g., people
with high verbal intelligence test scores)
show lower metabolic rates than low per¬
formers when attacking the same task.17 Evi¬
dently the harder the problem, the harder
the brain has to work, but the more intel¬
ligent the person, the less the brain has
to work. Intelligent people have efficient
brains.
16 Just, Carpenter, Keller, Eddy, et al., 1996.
17 See Neubauer, 2000, and Vernon, Wicket, Bazana,
& Stelmack, 2000, for reviews.
76
HUMAN INTELLIGENCE
3.10. A Critique of the Levels
Approach
Do we need three levels of theories?
There is a compelling argument for the
psychometric level. The tests are needed,
solely because of a requirement for the effi¬
cient measurement of intelligence in person¬
nel selection and utilization, ranging from
university entrance examinations to indus¬
trial employment programs. Since the tests
are needed, some orderly way of thinking
about the results is also required. Psycho¬
metric theories provide a way of doing this.
There is an equally compelling, but quite
different, argument for brain-level theories
of intelligence. They are an important step in
understanding, for they link complex behav¬
ioral observations, such as verbal compre¬
hension, to events inside the brain. Under¬
standing these links is an essential part of
the reductionist program of identifying the
causal links between individual differences
in brain processes and structures and indi¬
vidual differences in cognition.
Do we need a theory of intelligence at
the information-processing level, interven¬
ing between models of individual differences
in the brain and individual differences in test
performance and in daily life? I believe that
we do.
A complete information-processing
theory of intelligence would provide (a] an
understanding of the information¬
processing mechanisms that underlie
thought, (b) measurements of individual
differences in the power of these mecha¬
nisms, and (c) measurements of individual
differences in the knowledge bases that are
manipulated by the information-processing
mechanisms. I see this sort of theory as
a substantial advance on a theory that
explains the way intelligence is expressed
outside the laboratory in terms of corre¬
lations between measures of real-world
cognition and either test scores or statistical
summaries of them. I do not believe that
a model of intelligence based on the brain
can fulfill this role, because in practice we
are often interested in individual differences
in a capacity for processing information,
not individual differences in activation
levels in this or that brain region. For
instance, in assigning air traffic controllers it
is important to know how many different
aircraft a given controller can handle at one
time - and there are individual differences
in this ability. The requirement is stated
in terms of information-processing behav¬
iors. Psychological models to address this
sort of issue must be stated in terms of
information-processing capabilities.
This argument is not universally ac¬
cepted. Ian Deary, a professor at the Univer¬
sity of Edinburgh, has made a cogent argu¬
ment against information-processing models
of intelligence.18
Deary points out two weaknesses of
current information-processing approaches.
The first is that there is no guarantee that
the tasks used to measure information¬
processing functions, such as the span task,
measure what they say they do, because
there is no agreed-upon theory of cog¬
nition at the information-processing level,
and therefore no unassailable measure of
the postulated functions. The second weak¬
ness is that many of the alleged elementary
information-processing tasks require some
of the same behaviors as the psychometric
tasks. He does not see an advantage in restat¬
ing a psychometric theory in information¬
processing terms.
Deary proposes that research on intelli¬
gence ought to try to link psychometric test
scores directly to physiological measures.19
Deary says that his view of the current state
of theories of intelligence conforms to “the
Scottish stereotype of dourness; stern and
unrelenting.”20 I am somewhat more opti¬
mistic about the need to have information¬
processing levels of explanation, and a bit
more dour, stern, and unrelenting concern¬
ing the utility of studying a system consist¬
ing solely of brain-based and psychometric
variables.
The brain does not provide a place for a
complicated trait like intelligence ; it provides
18 Deary, 2000, Chapters 5-8.
19 Deary, 2000, Chaper 9.
20 Deary, 2000, p. 329.
ON THEORY
77
a kit of tools, the information-processing
functions, that, when guided by knowledge,
can be assembled into intelligent behav¬
ior. The way in which the brain is orga¬
nized, at the functional level, defines the
kit of tools and tells us how they must be
connected. Concepts like working memory,
attention, and speed of information pro¬
cessing are useful ways of talking about
the functional underpinnings of intellec¬
tual behavior, just as concepts like accel¬
eration, deceleration, gasoline consumption,
and turning ratio are useful ways of talking
about the functional underpinnings of auto¬
mobile capabilities.
The problem with going directly from
brain-based variables to psychometric vari¬
ables is that the steps are simply too com¬
plex to comprehend. While we now know a
great deal about the involvement of differ¬
ent regions of the brain in different types of
cognitive activity, we know much less about
how networks of neurons create memories
and conduct inferential reasoning, although
we have interesting mathematical models
of how neural networks might accomplish
these functions.21 Advances in the cognitive
neurosciences will tell us what the neural
bases of working memory, long-term mem¬
ory storage, attention, and similar functions
are. We will still need to understand how
these functions are assembled to produce
intelligent behavior. And we need to allow
for the possibility that different people will
assemble them in different ways.
I have somewhat more sympathy for
Deary’s “dour and unrelenting” view of cur¬
rent progress. He questions whether or not
the tasks that have so far been used to assess
information-processing concepts actually do
this, and do this at a level that is more
elementary than the behaviors required on
a psychometric test itself. This issue can
be resolved only on a case-by-case basis. It
is true that there are disagreements about
what the basic information-processing func¬
tions of the mind are, and about how each
21 This research is done under the general topic of
connectionist modeling. See E. Hunt (2002, 2007)
for discussions.
function should be measured. However, the
more research we do, the more psycholo¬
gists seem to be converging on an agreed-
upon information-processing model, includ¬
ing agreement on how its properties should
be measured and how these measurements
should be coordinated with information
about how the brain functions. Thus there is
no qualitative difference between my view
and Deary’s. It is just that we vary in opti¬
mism. Deary puts his eye squarely on the
empty portion of the glass, albeit admitting
that there is some liquid in it. My eye is more
focused on the liquid, but I certainly admit
that the glass is not full.
3.11. Summary
Theories are needed for two reasons - in
order to develop models for predicting and
controlling important systems and in order
to reduce complex phenomena to more
basic levels. Scientific theories are further
distinguished by a commitment to empiri¬
cism, objectivity, and empirical verification.
Theories of intelligence can be stated at
the psychometric, information-processing,
and biological levels. All are needed, but
for different purposes. Psychometric the¬
ories provide concise summaries of the
dimensions of variation in cognitive com¬
petence. They also play an important role
in personnel selection systems. Information¬
processing theories provide a functional
description of why the variables at the psy¬
chometric level correlate in the way they
do. This includes coordination with mea¬
sures of brain processes. Biological theories
relate intelligent behavior to individual dif¬
ferences in brain structure, brain processes,
genetic inheritance, or any combination of
these three.
Psychometric-level models and biological
level models relate observables to observ¬
ables. Information-processing models are, in
a sense, fictions. They are extremely useful
fictions. Indeed, they may be necessary. In
addition to being the appropriate level at
which to predict some aspects of socially
relevant behavior, information-processing
?8
HUMAN INTELLIGENCE
theories provide a vital link between the
psychometric and biological levels.
Cogent criticisms can be directed at all
current models at all levels. Criticizing cur¬
rent progress is not the same as denying
the need for models. We never know when
we have the “correct” model, for as we
push our observations further we are almost
certain either to change our current ideas
or to decide that they are accurate within
a particular set of observations, but need to
be accommodated to allow consideration of
other observations. This has been the nature
of progress in every other science; there is no
reason to expect it to be any different in the
scientific study of intelligence.
CHAPTER 4
Psychometric Theories
The definition of an ability arises from
systematic observations of individual
differences in performance on a defined set
of tasks. These observations constitute the
empirical basis of ability measurement.
They require no assumptions or exact
knowledge about neurophysiological
functions that might be responsible for
performance levels; although specialists
outside the strict field of psychometrics
may find it possible and useful to seek such
knowledge.
John B. Carroll, 1993, p. 23
Psychometric models of intelligence assume
that variations in cognitive performance
across different situations can be summa¬
rized by individual differences in a fairly
small number of basic cognitive dimensions,
such as general reasoning, verbal facility,
and visual-spatial reasoning. In general, the
effort to find these dimensions has been
quite successful. In the last analysis, though,
psychometric models are summaries of test
scores, and are relevant to cognition in gen¬
eral to the extent that test performance
itself is relevant to individual differences in
thinking, as revealed in daily life. In Chap¬
ter 1, I argued that test performance is rele¬
vant, as an imperfect sample of the cognitive
skills we use on a daily basis. This chapter
concentrates on theories of cognition within
that imperfect sample. The following chap¬
ter discusses expansion of these theories.
4.1. What Are Psychometric Models?
Whenever we summarize scores on differ¬
ent pieces of performance we are implic¬
itly assuming that there is some common
thread underlying each piece. This is a very
common assumption. The SAT math and
verbal scores are summaries of the scores
on subtests, and these summaries are them¬
selves amalgamated to produce an overall
score. Similar summaries are used by the
US Armed Services (summarizing sub tests
on the ASVAB to produce the AFQT),
and on what are avowedly intelligence tests,
like the WAIS (four index scores] and
the Woodcock-Johnson (WJ] tests (fluid
and crystallized intelligence, and a number
79
8o
HUMAN INTELLIGENCE
of other scales.) Psychometric theories are
basically expansions on this idea. Initially
tests are chosen to represent the variety of
cognitive performances that the investigator
is interested in. Statistical analyses, which
are reviewed briefly in the next section, are
used to discover the latent traits underlying
performance on ostensibly different tests. In
the case of research programs the analysis
often stops there. In the case of continu¬
ing programs of test development, such as
the SAT and WAIS programs, the analyses
of present tests are used to create improved
revisions of the original test.
The next section is a brief description of
factor analysis, a key statistical tool in test
development. Following sections discuss the
theories that have been developed using the
tool.
4.2. Factor Analysis
This section is intended to develop an intu¬
itive feeling for the method. It certainly is
not a substitute for a text on modern factor
analyts.
A psychometric model takes as the data
to be explained the covariances [or cor¬
relations, in the case of standard scoring)
among K observable measures. The expla¬
nation offered consists of a statement of
M underlying hypothetical variables, called
latent traits. The number of latent traits,
M, should be substantially smaller than the
number of tests, K. Further restrictions may
also be implied.
Now let us unpack these mathematical
ideas with a simple example.
4.2.1. The Introductory Example
Suppose that high school students were
given the following tests:
Vocabulary: A test of their ability to
define words.
Sentence comprehension : A test of their
ability to define the meaning of sen¬
tences.
Numerical computation : A test of their
ability to do computations involving
the basic operations of addition, sub¬
traction, multiplication, and division.
Algebra: A test of their ability to solve
algebraic problems.
Word problems: A test of their ability
to solve mathematical problems pre¬
sented in words. (According to one
cartoonist, the library in Hell contains
nothing but books of word problems.)
Our intuitions, which in this case would
be supported by a great deal of research,
are that in a typical class there would be a
strong tendency for the students who did
well on one test to do well on the others.
Conversely, students who did poorly on one
test would tend to do poorly on the oth¬
ers. This condition, in which all tests in a
battery are positively correlated, is called
positive manifold.
Positive manifold is a matter of degree,
for correlations between tests are seldom
perfect. The best student on the mathe¬
matics tests might not be the best student
on the language-related tests, but he or she
would probably be toward the top. The
same thing would be true of the worst stu¬
dent on the mathematics tests - probably
toward the bottom on the verbal tests but
not necessarily right at the bottom. To sum¬
marize, performance on one test would be
a useful predictor of performance on the
other tests, but the prediction would not be
perfect.
The hypothetical example reflects the
reality of intelligence testing. Virtually all
tests of cognitive performance exhibit posi¬
tive manifold to some degree, and they often
do so quite strongly.
In the hypothetical case, it seems reason¬
able to suppose that there are two under¬
lying abilities involved, verbal and math¬
ematical ability. If the two abilities are
themselves correlated, it might be appropri¬
ate to talk about a general cognitive ability.
But how are we to find out, in an objective
manner, whether we can reduce the five¬
dimensional space defined by the observable
test scores to just two or three dimensions
based on underlying traits? Factor analysis
answers this question.
PSYCHOMETRIC THEORIES
8l
Figure 4.1. A hypothetical example in which five tests are given to a
group of high school students. Each test can be thought of as an
arrow that points outward from the origin (which represents zeroes
on all tests). The more highly correlated the tests, the more their
vectors will point in the same direction. In this example the
mathematical and verbal tests form separate groups, and the word
problems test, which contains both mathematical and verbal
components, lies in-between the groups.
Factor analysis comes in two major vari¬
eties - exploratory factor analysis and con¬
firmatory factor analysis - and several sub-
varieties. First we consider the exploratory
technique.
4.2.2. Exploratory Factor Analysis
Here is a geometric way of thinking about
the five tests just introduced. Imagine that
each test is an arrow, pointing somewhere
in a space defining mental abilities, and
that each person's score is a point some¬
where along that arrow. If there is no cor¬
relation between any pair of tests, each of
these arrows will be placed at right angles
to [be orthogonal to) all of the others, so that
the arrows would point along the axes of a
five-dimensional space, and no further sim¬
plification would be possible. If the tests
are positively correlated, the arrows will
tend to point in the same general direction.
The higher the correlation between any two
tests, the more closely the arrows will point
in the same direction. In the extreme, if the
correlation between two tests is one, the cor¬
responding arrows will point in exactly the
same direction.
The situation is shown in Figure 4.1,
which is, of necessity, a two-dimensional
projection of the directions of the arrows
representing the tests. Exploratory factor
analysis finds a line that, in a precisely
defined mathematical sense (which need not
concern us, other than knowing that the def¬
inition exists), best summarizes the com¬
mon direction in which the tests are point¬
ing. This is shown by the dashed arrow in
Figure 4.2. It is called the first or general factor
for the five-test battery. This provides a con¬
siderable simplification of the data. Instead
of representing a person's score by five num¬
bers, for five tests, we can represent the score
by just one number, his or her factor score
on the general factor. Obviously some loss
of accuracy will result, just as a summary
score on a test loses information about what
problems a student did or did not answer
correctly. But how much accuracy has been
lost?
82
HUMAN INTELLIGENCE
Figure 4.2. The first and second factors extracted from the
hypothetical tests shown in Figure 4.1.
In order to answer this question we work
backward, by determining the correlations
between the first factor scores and each
of the original scores. These are called the
loadings of the original test on the first fac¬
tor. Unless the loading is 1 on every test
(which would imply that all the correlations
between tests were also 1) there will be some
loss of accuracy; the first factor score does
not predict the original test scores perfectly.
How great a problem is this?
Consider a particular student, Ignatz.
Ignatz's score on the general factor will be
influenced by his score on all five of the orig¬
inal tests. Suppose that Ignatz did excep¬
tionally well on the vocabulary test, but
poorly on the other tests. His score on the
general factor will therefore be lower than
would be expected from the vocabulary
test score alone. Factor analysis represents
Ignatz 's vocabulary test score by two scores,
the score predicted from Ignatz’s position
on the general factor and a residual, reflect¬
ing the difference between the actual score
and the predicted score.
Residual score on vocabulary
= Actual score - score predicted from
first factor
Actual score = Score predicted from
first factor + residual score
Repeating this procedure, the residual
scores can be correlated, the machinery of
exploratory factor analysis applied to the
residual correlations, and voila , we have
a second factor representing any common
trend in the “directions” defined by the resid¬
ual scores. However, these directions will be
lines in a four- dimensional space, because
one dimension has been “lost” to the first
factor. The result of this process is shown by
the second dashed line in Figure 4.2.
This process can be continued, analyz¬
ing residuals, then residuals of residuals,
until the size of the residuals becomes so
small that we may treat them as “noise.”
(When the number of factors extracted is
equal to the number of tests, the residuals
will be identically zero.) In most applica¬
tions the residuals become small enough to
ignore after only a few iterations. At that
point we stop extracting factors. More gen¬
erally, we say that we extract M factors from
K tests, where M is always less than K, and
in most cases substantially less. In our hypo¬
thetical example we stop at two factors, as
shown in Figure 4.2.
The next step is to develop a psychologi¬
cal interpretation of the factors. At this point
art supplements science.
In the example the first factor lies
between the cluster of verbal tests and the
cluster of mathematical tests. Let us call this
PSYCHOMETRIC THEORIES
83
Mathematical ability
Arithmetic
Figure 4.3. A correlated (oblique] factor solution for the
hypothetical example, defining correlated verbal and mathematical
abilities.
factor “general competence.'’ The direction
of the general factor is very close to the
direction of the arrow representing the word
problem test. This is reasonable, because
solving word problems requires a combina¬
tion of verbal and mathematical skills. The
word problem test would be called a marker
for the general competence factor, because
the test and the factor point in nearly the
same direction.
Because the second factor summarizes
residual scores, the first and second factors
must be uncorrelated. Geometrically, the
arrows for the two factors will be at right
angles to ( orthogonal to) each other. The two
verbal tests and the two mathematical tests
point in different directions along the sec¬
ond factor. Mathematically, we would say
that one set of tests has a positive loading
and the other has a negative loading on the
second factor.
One interpretation of the second factor in
our hypothetical example is that it reflects
negative correlations between the residuals
of scores on the language and mathematical
tests, suggesting that people can be classi¬
fied as “relatively verbally skilled" or “rel¬
atively mathematically skilled,” after hairing
taken account of their general competence. The
qualifying phrase is important. Findings like
those in this example would not be evidence
that people are either “good at math" or
“good at language tasks,” but not both. It is
only after the effects of general ability have
been removed that a tilt toward linguistic or
mathematical problem solving appears.
This example is not completely made up.
It is a simplification of an analysis conducted
by Werner Wittmann, at the University of
Mannheim in Germany. Wittmann analyzed
scores achieved by fifteen-year-old students
on the Program for International Student
Assessment (PISA) mathematics and sci¬
ence examinations.1 His two-factor solution
resembles the one offered here.
The division of the two factors into a gen¬
eral factor and an orthogonal residual has
been dictated by the requirement that the
two factors be orthogonal to each other.
While this simplifies the mathematical pre¬
sentation, there is no compelling argument
that a mathematically simple explanation
will always lead to the clearest psycholog¬
ical explanation. An alternative way of pro¬
ceeding is to use factor analysis to identify
the number of dimensions required, two in
this case, and then to rotate the dimensions
so that each of the factors lies amid a clus¬
ter of similar tests. (These notions can be
given a mathematical definition.) Figure 4.3
shows how this method of factor defini¬
tion would be applied to the hypothetical
1 Wittmann, 2005.
84
HUMAN INTELLIGENCE
example. This model identifies two psycho¬
logically interpretable factors: mathemati¬
cal ability, with high loadings for the two
overtly mathematical tests, and verbal abil¬
ity, with high loadings on the two verbal
tests. Instead of regarding the word prob¬
lems test as a marker for a general compe¬
tence factor, it is seen as a test requiring an
amalgam of mathematical and verbal ability.
This avoids the problem of dealing with a
factor that seems to demand a negative cor¬
relation between abilities. The simplification
has come at a cost. The mathematical model
has been made more complex by adding an
additional parameter, showing the correla¬
tion between factors.
In fact, one could argue that the fact that
the mathematical and verbal ability factors
are correlated is itself evidence for a general
factor.
This example illustrates both the power
and the weakness of exploratory factor anal¬
ysis. The technique provides an objective
and accurate technique for determining how
many factors are required to account for
individual variation in the scores, across
tests. Assigning meaning to the factors
requires a mix of further mathematical
assumptions and intuitions about the psy¬
chology underlying the variation. Given this
mixture of mathematics and psychology, it
is not surprising that a good deal of the liter¬
ature on intelligence from the 1930s through
the 1960s reflected debates about how fac¬
tor analyses should be interpreted. The field
was considerably advanced by the develop¬
ment of confirmatory factor analysis in the
1970s.
4.2.3. Confirmatory Factor Analysis
Confirmatory factor analysis provides a way
of deciding how well a theoretical model
matches the observations in a psychometric
study.2 The investigator decides, in advance,
the relationships that he or she believes
hold between factors and tests, and for that
2 Confirmatory factor analysis is a subset of a more
general statistical technique, structural equation
modeling. As this method is fairly complicated I will
touch on it only when necessary.
matter, between the factors. Computation¬
ally demanding algorithms are used to com¬
pare the observed relations between mea¬
sures to the investigator’s assumptions. This
moves factor analysis from a method for
inducing relations from data to a method
for testing hypotheses about data - a prefer¬
able approach in data analysis. The follow¬
ing example illustrates the method. The
example is due to Karl Joreskog, a Swedish
psychometrician who was one of the major
developers of confirmatory factor analysis,
and his colleague Dag Sorbom.5
Psychological tests can be given in either
speeded or unspeeded conditions. In a
speeded test examinees have to answer all
questions within a fixed time period; in
an unspeeded test examinees take as much
time as they like. Joreskog used confirmatory
factor analysis to investigate the question
of whether speed pressures influence some
people more than others. The examinees in
this study were given four vocabulary tests.
Two of the tests, %, and x2, were fifteen-
item tests, administered with ample time to
complete all fifteen items. The other two
measures, yl and y2, were seventy-five-item
tests, administered under strict, tight time
limits. The question of interest is whether
the same trait, word knowledge, was mea¬
sured in each condition, or, as is some¬
times asserted, there are some people who
just do not do well on speeded tests, even
if they know the information? Table 4.1
shows the variance-covariance matrix for
the four tests, which is what was analyzed.
The bottom part of the table shows the cor¬
relation coefficients implied by the variance-
covariance terms.
Joreskog and Sorbom assumed that dif¬
ferent, possibly correlated latent traits
( factors ) underlie performance on the two
types of vocabulary tests. Their assumption
is shown in Figure 4.4, using the graphic con¬
ventions introduced in Chapter 1. Causative
relationships are indicated by single-headed
arrows. For example, the first factor, f, is
assumed to be a cause of performance on the
3 Joreskog & Sorbom, 1979. The example is discussed
on p. 52 ff. The original data is due to Lord (1956).
PSYCHOMETRIC THEORIES
85
Table 4.1. The variance/covariance and correlation matrices
in the confirmatory factor analysis example. Tests x, and x2
are unspeeded vocabulary tests; tests y, and y2 are
vocabulary tests completed under a stringent time limit
Variance/Covariance Matrix xi X2 yi y2
XI
86.3979
X2
57-7751
86.2632
yi
56.8651
59-3177
97.285
y*
58.8986
59.6683
73.8201
97.8192
Correlation Matrix
XI
X2
yi
y^
XI
1.00
X2
0.6r-
1.00
yi
O.62
p
ON
VJ1
1.00
y2
0.64
0.65
p
Ti
Os
1.00
Source: Data from
amalgamating information in
Tables 2
and 3 of
Joreskog & Sorbom, 1979.
two unspeeded tests. Accordingly there are
single-headed arrows from latent variable /,
to observable variables x, and x2 .
The boxes, ellipses, and arrows establish
the structure of a factor analytic model. The
e’s and the Greek letters indicate the three
types of parameters to be estimated. The
P variables indicate the relationship bet¬
ween the latent and the observed vari¬
ables. These represent causation; the latent
variable is seen as being one of the causes
of the observed variable, and the value of /3
indicates the extent of the causal relation.
The e variables represent residuals.
Residuals are not “error" terms in the
sense that an error is a mistake. They
represent the effects of all unmeasured fac¬
tors that might influence the observable
scores, but are not part of the model. In this
example each residual term is uniquely
> .
. " . *C±,>
£12
Pl^/^\^2A
\
rfi
| yi £3 n *~eA
Figure 4.4. The two-factor model used by Joreskog and Sorbom to
summarize correlations between unspeeded {x}, x2 ) and speeded
(ylf y2 ) vocabulary tests. The model assumes two latent traits, one
representing the ability to do speeded tests and another representing
the ability to do unspeeded tests. A correlation between the two
latent traits is permitted, with the correlation, p, to be estimated
from the data. The observed correlations are indicated by the lower
dashed lines between the x and y variables. This model could be
contrasted to a single-factor model by requiring that p be identically
one. Note the residual terms showing external influences on each of
the four tests. (Figure modeled after Joreskog & Sorbom, 1979,
Figure 3, p. 54.)
86
HUMAN INTELLIGENCE
associated with a measurable variable. It is
possible to construct models that postulate
an unmeasured residual variable that influ¬
ences several latent variables. If the system
of variables were to be completely closed, so
that the hypothesized factors were the only
factors influencing the observable measures,
the residuals would always be zero. In the
social and behavioral sciences this virtually
never happens. By examining residuals we
can get an idea of the extent to which a
model captures the variation between the
observables. The smaller the residuals, the
more the model captures the variations that
are to be explained.
The p symbols indicate the magnitude of
a correlation between variables, without any
implication of causation. Therefore, the p
symbols are always associated with double¬
headed arrows.
The absence of a link between two vari¬
ables indicates that the model being tested
assumes that the two variables are not
linked, that is, that the appropriate or p is
identically zero.
This model makes strong predictions
about the observed correlations among the
four tests. Consider the two unspeeded tests,
the x variables. If the model is accurate,
the covariance between them, Cov (x1; x2),
should depend only on (a) their respective
variances and (b) the two parameters
1 321 . A similar argument holds for the y vari¬
ables. But consider the case of the covari¬
ance between an unspeeded and a speeded
test, for example, Xj andy2. The covariance
should depend on the test variances and
three parameters, j3u, p, and /332. Generaliz¬
ing, any assignment of values to the param¬
eters will imply predicted covariances (and
therefore correlations) between all pairs of
observed tests. These form the predicted
covariance matrix , E . It can be compared to
the observed covariance matrix, S. What con¬
firmatory factor analysis does is to find the
set of values for parameters that brings E
as close as possible to S. This is exactly the
same logic as is used, for instance, to esti¬
mate the mean and variance of a population
from a sample. Only the details are more
complicated.
4.2.4. Statistical Evaluation
and Model Comparison
Let M be a collection of parameters, some
of which have been estimated and some of
which have been set by theory. We will refer
to this as a model. Figure 4.4 is an example
of a model. Any model M implies a pre¬
dicted covariance matrix, E(M), that can
be compared to the observed covariance
matrix, S. Statistical inference techniques
can be applied to determine the extent to
which the two agree. This is called the
fit of the model to the data. Evaluations
of fit can include determining whether or
not the differences are greater than would
be expected “by chance.” This is logically
identical to testing a null hypothesis, as
described in introductory statistics courses.
(The computations are quite a bit more
complicated!)
It is often possible to go beyond evaluat¬
ing fit. We can compare two different mod¬
els, Mi and M2, to each other. The simplest,
and often the most useful, case is one in
which the parameters to be estimated in M2
are a subset of those to be estimated in Mi.
This can be illustrated using the model in
Figure 4.4, which will serve as Mi.
Mi assumes that two ability factors are
involved, the ability to take speeded vocab¬
ulary tests and the ability to take unspeeded
ones. However, the model allows for the
(reasonable) possibility that the two may
be correlated, that is, that people who do
well on unspeeded tests tend to do well
on speeded tests, but that the two talents
are not exactly the same. The extent of the
correlation is indicated by the p parameter,
whose value is to be estimated from the data.
A fairly obvious alternative is to assume that
all vocabulary tests tap the same underlying
trait, strength of vocabulary, and that it does
not matter if the tests are speeded or not. If
that is the case, Figure 4.4 should be modi¬
fied to show just one latent trait, /, instead
of/j and/7. Call this model M2.
Here is an important insight. Saying that
there is one trait is equivalent to saying that
there are two traits but that they are per¬
fectly correlated, p — 1. Therefore, M2 has
PSYCHOMETRIC THEORIES
87
the same parameters as Mi, but, and it is
a very important “but,” in M2 p is fixed at
1 instead of being estimated from the data.
This has two important consequences. The
first is that M2 cannot have a better fit than
Mi, because the computational techniques
have ensured that in Mi p has the best pos¬
sible value it can have in order to get a good
fit to the data. The second consequence is
that we can compare the fits of the two
models, to determine whether the differ¬
ence between fits shows that Mi is more
accurate than M2 than would be expected
by chance. Going back to the example, we
can determine whether the data requires
that we assume separate vocabulary skills
for speeded and unspeeded tests, or whether
it is possible to make the simpler assump¬
tion that all the tests evaluate a single trait,
vocabulary knowledge.
This illustrates a very important advance
in theory construction. Confirmatory factor
analysis permits testing of the relative accu¬
racy of different models as summaries of the
same data.
In this and subsequent chapters the
reader will be presented with several mod¬
els of the sort just described. The models
will typically involve many more observable
and latent variables, and the structure of the
models will be more complex than in the
simple example, but the principle of anal¬
ysis will be the same. Confirmatory factor
analysis allows us to establish a competi¬
tion between models, and this is how science
proceeds.4
4.2.5. Limits on the Generality
of Factor Analysis
The development of modern factor analy¬
ses has greatly enhanced the precision of
psychometric research. This point has not
always been appreciated by critics, who
sometimes appear to be attacking the field
for deficiencies in techniques that were
4 See Bouchard (2009). I have skipped over a num¬
ber of technical details. It is possible, although rare
in practice, to find situations in which alternative
solutions have such close fits that confirmatory fac¬
tor analysis cannot distinguish between them.
abandoned years ago.5 Nonetheless, there
are some limitations on generalizing the
results of factor analytic research.
Factor analytic studies are often used to
reveal the nature of underlying, unobserv¬
able traits, especially general intelligence.
Tests with loadings close to 1 are impor¬
tant pieces of evidence, because a loading of
l indicates that variation in observable test
performance is perfectly linked to variation
in the trait. For instance, the argument that
the ability to detect patterns is an important
part of general intelligence is supported by
the fact that progressive matrix tests often
have high loadings on the g factor, and that
pattern detection is an important step in
solving progressive matrix problems. In this
case, we say that progressive matrix tests are
markers for general intelligence.6
Unfortunately, it is easy to misinterpret
this statement. A high loading between a
trait and a test does not mean that the qual¬
ity indicated by the trait is required for
performance on the test in the way that,
say, strength is required to push a piano
across your living room. It means that varia¬
tion in test performance, across people, is
related to the way the trait varies, across
people. Loadings are statements about indi¬
vidual differences as observed in a partic¬
ular group. A person’s standings on both
trait and test are defined relative to the
performance of other individuals in the
group, not by the individual’s absolute per¬
formance. The mathematics whiz in high
school may not be that outstanding at MIT.
This has an important implication for factor
analysis.
The test loadings derived from factor anal¬
ysis result from an interaction between the
properties of the test itself, the other tests in
the battery, and distribution of traits in the
population being tested .
5 This is especially true of widely cited attacks by
the evolutionary biologist Stephen J. Gould (1981,
pp. 252-255). Surprisingly, the error was repeated in
a revised edition of this work (Gould, 1996) even
though the error had been commented upon by
reviewers of the original work.
6 Jensen, 1998.
88
HUMAN INTELLIGENCE
Some thought experiments illustrate
what this means.
Here is a list of nine possible tests:
1. A vocabulary test.
2. A test of paragraph comprehension
based on newspaper editorials.
3. A test requiring judgments of the syntax
of English sentences.
4. A test requiring people to evaluate the
logical argument contained in a political
speech.
5. A test in which people are asked to
judge whether or not meaningless pat¬
terns presented at different orientations
are identical or whether one is the mir¬
ror image of another. This means that
the examinee has to rotate figures “in
the mind's eye.” For this reason such
tests are called rotation tests.
6. A test in which people view a room
(e.g., an office) and then imagine that
they are at one location, facing in a par¬
ticular direction, and point to a target
object. For example, “imagine that you
are at the desk, facing the file cabinet.
Point to the flower pot.”
7. A test in which the task is to find a
picture hidden in another; for exam¬
ple, finding a triangle in a Star of David
figure.
8. A map-reading test.
9. A word problem test in which peo¬
ple read passages about an explorer
going through a jungle and then answer
questions about the spatial layout of
objects. To illustrate: “As Henry pro¬
ceeded southward, he realized that the
python was keeping up with him on the
right and the tiger on the left.” Ques¬
tion: Was the python west of the tiger?
The first four tests evaluate language skill;
the next four evaluate visual-spatial reason¬
ing; and the last test evaluates both language
skills and visual-spatial reasoning. Suppose
that these tests have been constructed so
that the scores are normally distributed in
the high school population. Because visual-
spatial and language skills are correlated in
the high school population, it would be
possible to extract a general factor and resid¬
ual spatial and verbal factors.
Imagine two further studies. In the first
we give the test battery to a representative
sample of lawyers. In the second it is given to
a representative sample of architects. What
would happen to the factor loadings? I sug¬
gest the reader stop and think a moment.
Lawyers are highly selected for language-
related skills. Therefore, the four tests of
language skills, although appropriate for a
high school sample, would be quite easy for
the lawyers. As a result there would be lit¬
tle variation in scores on the verbal tests.
Lawyers are not selected for their visual-
spatial skills, so we might find considerable
variation on the spatial tests. An analysis
for a single general factor would result in
a factor with high loadings on spatial tests,
because that is where the variance would be.
Test 9, which draws on both verbal and spa¬
tial skills, would have a high loading on the
spatial residual factor and a low loading on
the verbal factor.
The spatial skills tests would be easy for
the architects, while tests of verbal skills
should show more variation. Accordingly,
it is likely that the general factor, as defined
for architects, would have high loadings on
verbal tests. Test 9 would now have a high
verbal loading and a low spatial loading.
These changes in loading have nothing to
do with the absolute abilities of the indi¬
viduals involved. They are driven by the
amount of variation in the population, not
the level of skill. The examples are not con¬
trived; they represent a situation that occurs
often in the literature. It is common prac¬
tice to conduct psychological studies using
college students as participants. College stu¬
dents are selected largely on verbal skills
and on general intelligence. As a result, the
variation among college students on these
traits is less than the variation in the popu¬
lation. Therefore, a study of the importance
of intelligence in college students is likely
to underestimate the importance of intelli¬
gence (or verbal skills) in the population at
large.
The next thought experiment deals with
another problem that limits the generality
PSYCHOMETRIC THEORIES
89
of factor analytic findings. The loading of
a test on a trait can vary depending upon
what other tests there are in the battery,
even though the same population is studied.
Suppose that instead of using all nine
of our tests we compose a smaller battery
consisting of the first five tests - four tests
of language skills and the rotation test. The
test battery is given to a representative set
of adults. We then extract a single general
factor from the data. It would be defined by
the traits required by the four verbal tests.
The single spatial test (test 5) would have a
low loading on the general factor and a large
residual term. The vocabulary test (test 1]
would have a high loading on the general
factor, for it has repeatedly been found that
people with high general verbal skills tend
to have large vocabularies.
Now construct a new battery consisting
of the vocabulary test and the four visual-
spatial tasks, tests 5-8. The general factor will
be determined largely by the visual-spatial
tasks. The rotation test would have a high
loading on this factor, while the vocabulary
test would have a low loading and a high
residual term.
These two thought experiments show
what might happen, not what does. In prac¬
tice, investigators use similar test batteries,
which leads to replicability of results, but at
the same time produces a restricted defini¬
tion of intelligence, because the same mental
skills are being evaluated from experiment
to experiment, in only slightly different
ways. Factor structures for a given battery
are usually replicated across populations
(college students, military recruits, popula¬
tion samples of adults), although there are
occasional failures to replicate. The restric¬
tions on generality developed here are warn¬
ings to keep in mind, not compelling argu¬
ments for throwing out all factor analytic
studies.
4.3. The Theory of General
Intelligence (g)
The general intelligence (g) theory is at once
the best-known, most praised, and most
vilified psychometric model of intelligence.
The theory asserts that individual differ¬
ences in cognitive performance are very
largely due to individual differences along
a single dimension of mental competence,
general intelligence (g). Psychologists who
advocate the g model acknowledge that
there are other dimensions of mental per¬
formance, such as verbal or visual-spatial
reasoning, but claim that these dimensions
account for much less of the variation in
cognitive performance than the g dimension
does.
4.3.1. Early Development of the Theory
The g model was first stated by Charles
Spearman, an Englishman who has some
claim to be the first modern psycho¬
metrician.7 After serving for some time as an
army officer, Spearman entered academia,
receiving a Ph.D. from Wilhelm Wundt’s
experimental psychology laboratory. He was
attracted to Galton’s concept of intelli¬
gence as generalized mental fitness, but did
not share Galton’s enthusiasm for what we
would, today, call information-processing
measures of cognition. Instead he thought
that one should find a general factor through
the analysis of complex measures of think¬
ing, such as school grades. In order to inves¬
tigate his ideas he developed both the basis
for early factor analysis and the rank-order
correlation coefficient, thus making substan¬
tial contributions to the budding science of
statistics. David Wechsler, the developer of
the WAIS, was a student of Spearman’s, as
was Raymond Cattell, whose work will be
described later in this chapter. His work also
heavily influenced the ideas of Philip E. Ver¬
non, Hans Eysenck, Cyril Burt, and Arthur
Jensen, all of whom have made major con¬
tributions to the field of intelligence.
Table 4.2 presents a table of correlations
between a set of school grades that Spear¬
man used as evidence for g. Following Spear¬
man’s practice, the classes are arranged in
order of the average correlation between one
grade and the other grades. To illustrate,
7 Spearman, 1904, 1923, 1927.
9o
HUMAN INTELLIGENCE
Table 4.2. The correlations between school grades used by Spearman as an illustration of
the theory of general intelligence (g). The figures in parentheses are the predicted
correlations when Carroll (1993) applied the theory using modern computing methods.
Class
Classics
French
English
Mathematics
Pitch
Music
Classics
French
English
Mathematics
Pitch
Music
•8? (.84)
.76 (.77)
.70 (.72)
.66 (.64)
.63 (.62)
1
.67 (.70
.67 [.66]
.65 (.60)
•57 (-57)
1
.64 [.60]
■54 ('54)
■5i (-52)
1
•45 (-5°)
•51 C-48)
1
.40 (.43)
\
Classics, the first grade on the table, has the
highest average correlation with the other
grades.
Spearman argued that a student’s grade
in a class could be thought of as a mea¬
sure of the sum of a student's general¬
ized intelligence and the student’s unique
abilities and knowledge relevant to the class.
Algebraically this is
Xij = Pjgi +Sij, [4-0
where %ij is person is grade on topic j,
gi is person is general intelligence, and sq
is person is ability on a trait that reflects
those aspects of test j that are not associated
with general intelligence, is a term that
reflects the importance of the general factor
in determining the grade in the ;th class. In
modern terms, is the factor loading of test
j on the general factor, and 5,9 is the residual
term for person i on test j.
There were no computers in Spearman’s
day, so he used clever but, by today's stan¬
dards, archaic methods of estimating factor
loadings. Almost one hundred years later a
modern psychometrician, John Carroll, fac¬
tor analyzed Spearman’s data.8 The resulting
model is shown in Figure 4.5. The parenthe¬
sized terms in Table 4.2 are the correlations
that would be expected if the g model were
an accurate summary of the data. As can
be seen, the observed and predicted corre¬
lations [the S and T matrices of section 4.2)
are quite close. Spearman’s theory is clearly
supported by this data.
8 Carroll, 1993, p. 38.
Although Spearman and his modern fol¬
lowers emphasized the importance of g,
they recognized the existence of group fac¬
tors, intellectual skills, such as facility with
language or facility in dealing with visual-
spatial patterns, that are less general than g
but more general than an ability to take a
specific test. Figure 4.6 shows a proposed
analysis of our eight hypothetical tests in
terms of Spearman’s expanded model. All
tests would be expected to load on the g
factor, but the verbal tests would also be
expected to load on a language skill factor,
and the spatial tests would be expected to
load on a visual-spatial factor.
In Figure 4.6 the g, language factor, and
the visual-spatial skill factors are not allowed
to correlate with each other or with g.
[This is indicated by there not being any
line connecting the three factors.] Spear¬
man had to make this assumption because
he did not have access to the computational
power required to evaluate more compli¬
cated models. Modern computing allows us
to consider a more sophisticated version,
the hierarchical model shown in Figure 4.7.
In this model Spearman’s group factors are
referred to as (broad) second-order factors,
and g, a third-order factor, is inferred from
the correlations between the second-order
factors. We will see several other examples
of hierarchical models subsequently.
4.3.2. The Evidence for g
Spearman’s model has lasted longer than
most psychological theories do. Arthur
Jensen, a prominent modern advocate of
PSYCHOMETRIC THEORIES
91
.763
.740
.287
i
.471
.600
.661
Figure 4.5. Loadings and residual terms when the g model was
applied to Spearman’s school grade data. Loadings calculated by
Carroll (1993), residuals calculated by the author.
g theory, has claimed that after literally a
century of exploration the g model provides
a simple, accurate summary of a massive
number of studies of intelligence.9 Jensen
built his case on three lines of evidence.
The first is the widespread observance of
positive manifold. Virtually all tests of cog¬
nition are positively correlated. The second
line of evidence is that measurements of gen¬
eral intelligence are among the best predic¬
tors of performance both in school and in
the workplace. We will expand on this point
in Chapter 10. The third line of evidence
is that measures of g are related to a rel¬
atively small set of information-processing
functions and, perhaps more importantly,
to certain brain-based and genetic mea¬
sures. This point is discussed in Chapter 6.
Here we look only at the psychometric
evidence.
Positive manifold is a widespread statis¬
tical phenomenon. The following examples
show its ubiquity.
Gilles Gignac, an Australian psychome¬
trician, analyzed four widely used American
tests, the KAIT, the WAIS-R and WAIS-
III, and a multidimensional aptitude battery
(MAB) designed to cover a wide variety of
9 Jensen, 1998.
cognitive skills.10 While all analyses showed
substantially the same results, the WAIS-III
analysis is of particular interest because the
standardization sample was constructed to
represent the US adult population, using a
sample of just under 2,500 people ranging in
age from sixteen to eighty- nine.
Every subtest of the WAIS-III had sub¬
stantial loadings on the general factor. The
two highest loadings were on the Arithmetic
(.79] and Vocabulary (.75) subtests.11 This
certainly fits the definition of a general fac¬
tor, applying to a wide variety of cognitively
demanding tasks. The lowest loadings were
found on the Digit Symbol and Coding sub¬
tests [.53 and .56), which are essentially tests
of short-term memory. A matrix test simi¬
lar to the RPM described in Chapter 2 had a
loading of .70, which is consistent with the
argument that progressive matrix tests are
good markers for general intelligence.12
In addition to being ubiquitous, the g fac¬
tor was substantial. In Gignac’s study the
general factor accounted for from 30% to
10 Gignac, 2006. Gignac’s model somewhat resembled
Spearman's model (Figure 4.6], but the treatment
of residual effects was much more sophisticated.
11 See Chapter 2 for a description of WAIS-III subtests.
12 Jensen, 1998, pp. 37-38.
92
HUMAN INTELLIGENCE
Figure 4.6. Spearman’s g model expanded to allow for group
factors. In this hypothetical example the battery consists of eight
tests. All eight have loadings on the g factor. In addition, the four
tests involving language have a loading on a language-skill group
factor, while the four tests involving visual-spatial reasoning have a
loading on a visual-spatial factor. The two group factors are
uncorrelated.
60% of the variance in scores on individual
tests.
The American results are mirrored by
findings in other developed countries. Deary
and his colleagues analyzed the results
from a study in which the Cognitive Abil¬
ities Test was given to virtually all mid¬
dle school children in England.13 This test
was designed to provide separate scales for
verbal, quantitative, and nonverbal reason¬
ing, but the scales are not necessarily sta¬
tistically independent.14 Indeed they were
not. The general factor accounted for 70%
13 Deary, Strand, Smith, & Fernandes, 2007.
14 See Chapter 2 for a description of this test.
of the variance. In another British study
Elliott15 examined results from examination
of schoolchildren of varying ages, using the
British Ability Scales. Elliott’s study was a
challenging test of the g model, because the
British Ability Scales were designed to pro¬
vide clinical and educational psychologists
with a more differentiated view of cogni¬
tive abilities than that provided by g-loaded
tests, such as the WAIS. Nevertheless, the
general factor accounted for 30% of the
variance on these scales, even though they
had been designed to deemphasize general
intelligence.
15 Elliott, 1986.
PSYCHOMETRIC THEORIES
93
Figure 4.7. The example in Figure 4.6 reformulated as a
hierarchical model. The two group factors are now partially caused
by g, and thus would be correlated. This is the most widely
accepted model of intelligence today.
Roberto Colom and his colleagues at the
Universidad Autonoma in Madrid have ana¬
lyzed the normative data from the Spanish
version of the WAIS-III, doing separate
analyses for groups with different levels of
education.16 The g factor accounted for 45%
of the variance in the two least-educated
groups, and 25% of the variance in the
two better-educated groups. This turns out
to be a typical finding. The general fac¬
tor is often less powerful as cognitive skills
increase.
Psychologists from the University of
Tilburg in the Netherlands administered
three different cognitive batteries to groups
16 Colom, Abad, Garcia, & Juan-Espinosa, 2002.
of immigrant and native Dutch students.17
Depending on the battery, the g factor
accounted for from 30% to 45% of the vari¬
ance. The g-loadings of the various sub¬
tests were nearly identical in both the native
Dutch and the immigrant groups. This sug¬
gests a similarity of cognitive structures over
two different ethnic groups.
Similar results have been obtained in
many studies. In 1993 John Carroll pub¬
lished an extensive survey of the results from
many of the studies of batteries of tests that
had been conducted up to that time.18 He
found evidence for a g factor in 146 different
17 Helms-Lorenz, van de Vijver, & Poortinga, 2003.
18 Carroll, 1993.
94
HUMAN INTELLIGENCE
data sets. There appears to be little doubt
that analysis of virtually any battery of tests
of cognition, in any population within an
industrial society, will reveal a g factor.
As the thought experiments in section 4.2
showed, demonstrations of positive mani¬
fold in different populations and with differ¬
ent test batteries do not show that the dif¬
ferent studies have found the same g. Carroll
observed that in order to reach such a con¬
clusion one would have to administer several
test batteries to the same individuals, extract
the g factor from each battery, and show that
they were highly correlated.19 Subsequently,
Wendy Johnson and her colleagues at the
University of Minnesota produced just such
a finding. They administered three different
test batteries to a group of just under 500
adults, of widely varying ages.20 Their results
strongly supported the contention that the
g that is extracted by statistical analyses of
widely used test batteries is the same trait,
psychologically, regardless of the battery.
Because the Johnson et al. test provided
such strong evidence for a general reasoning
factor as both a statistical and a psycholog¬
ical phenomenon, we need to take a look
at the study itself. The batteries they used
were the WAIS and two batteries that had
been designed to sample a comprehensive
range of verbal, quantitative, and nonverbal
reasoning abilities. Their sample consisted
of adults from their twenties to their seven¬
ties who were participating in an extensive
behavior-genetic study of adoptees. The par¬
ticipants were generally white, of middle or
higher socioeconomic class, and from either
a North American or a European coun¬
try. Such a sample is certainly not statisti¬
cally representative of the population of any
country, let alone the world, but it is fairly
representative of an important segment of
the population within the industrial/post¬
industrial nations.
There is a massive amount of data show¬
ing a simple, clear-cut, and important fact.
Within the culture of the industrial/post¬
industrial world, which is where these
19 Ibid., p. 596.
20 Johnson et al., 2004.
studies were conducted, people can be reli¬
ably classified by the extent to which they
display general intelligence Q g ), that is, a
tendency to be cognitively competent in a
wide variety of tasks. Anyone who asserts
that individuals have to be described by sev¬
eral relatively independent cognitive traits is
simply wrong. A person who is very good
at verbal reasoning may not be the best at
quantitative reasoning or reasoning about
visual displays, but he or she is unlikely to
be poor at these other tasks. The same argu¬
ment applies, of course, to people who are
very good at quantitative or visual reasoning.
Their verbal reasoning may not be as good
as their reasoning in their strongest domain,
but it is unlikely to be bad.
4.3.3. What Is the Nature of g?
Positive manifold is a fact. There is no argu¬
ing with the existence of g as a statistical
phenomenon. There is a great deal of argu¬
ment over the reason that different evalua¬
tions of cognitive performance are virtually
always positively correlated.
One possibility is that there is a mental
trait, cognitive strength, analogous to physi¬
cal strength, that people possess to different
degrees and that, as is the case for phys¬
ical strength, mental strength is useful in
many settings. Let us call this the unitary g
hypothesis. It implies that tests of cognitive
competence will be positively correlated, as
they are. But this is not conclusive evidence.
Abstractly, if A (unitary g) implies B (pos¬
itive manifold), and B is observed, then A
might be the case, but B might have arisen
for other reasons. What hypotheses other
than the unitary g hypothesis imply positive
manifold?
Positive manifold would arise if people
possess distinct, rather specialized mental
traits, such as verbal and visual-spatial rea¬
soning, and these traits are correlated in
the population, for either environmental or
genetic reasons. Call this the correlated traits
hypothesis. Correlated traits would be anal¬
ogous to the statistical association between
blond hair and blue eyes, neither of which
causes the other.
PSYCHOMETRIC THEORIES
95
A third possibility is that there are sep¬
arate cognitive traits but that over the life
span there are positive interactions among
these traits, so that high or low performance
in one trait affects the development of oth¬
ers. The interaction could be due to biol¬
ogy, the environment, or some combina¬
tion of the two. For example, children who
appear to be highly verbal may be singled
out for special instruction, which improves
their cognitive capacity, opening yet further
opportunities. The opposite may also hap¬
pen; children who are perceived as being
slow speakers or readers could be treated as
if they were not too bright in general, lead¬
ing to a self-fulfilling prophecy of failure.
The first position, that g exists as a
trait, is the one held by most advocates of
the general intelligence model.21 But what
is the trait behind the statistical abstrac¬
tion? To quote from a statement by a
US president, "it depends on what you
mean by ‘is’,”22 for the question can be
answered at several levels. At the psycho¬
metric level general intelligence could be
defined by examining the marker tests that
have high loadings on the general factor,
and attempting to identify the common
cognitive challenges they present. At the
information-processing level general intel¬
ligence could be defined by showing that
the general factor is highly correlated with
particular information-processing functions,
such as working memory, the ability to
control attention, and general speediness.
At the biological level general intelligence
could be associated with individual differ¬
ences in brain structures and processes, and
with variations in the genome. All three
approaches have been tried. We will look
at the psychometric evidence in this chap¬
ter, saving the information-processing and
biological evidence for later chapters.
Spearman's belief that the ability to
detect patterns is central to intelligence led
21 See, e.g., Gottfredson, 1997; Jensen, 1998, 2006.
22 William (Bill) Clinton, commenting on the truth
of his lawyer’s statement that “there is absolutely
no sex of any kind” in Clinton’s relationship with
a White House intern, Monica Lewinsky (Shapiro,
2006, p. 160).
his student John Raven to develop the Raven
Progressive Matrices (RPM) tests described
in Chapter 2. The effort was successful, for
these tests do have high loadings on the gen¬
eral intelligence factor. Jensen states:
When the Progressive Matrices test is fac¬
tor analyzed along with a variety of other
tests it is typically among the two or three
tests having the highest g loadings, usually
around .80. Probably its most distinct fea¬
ture is its very low loadings on any factor
other than g.
Jensen, 1998, p. 38
This is what would be expected if Spear¬
man were right, for a person who takes an
RPM test does have to be able to notice
visual patterns. However, the matter is not
quite as clear as Jensen's statement sug¬
gests. Other tests, which appear to call forth
very different sorts of problem solving than
that required on the RPM tests, also have
high loadings on g. In particular, vocabu¬
lary tests have nearly as high loadings on
g as do RPM tests, a finding that has been
found in samples as diverse as the WAIS-III
normalization sample (representative of the
US population),25 Swedish schoolchildren,^
and a large group of adults who volunteered
for a research project.25
After reviewing findings from a great
number of studies, Carroll26 found that four
different classes of tests were consistently
found to have substantial loadings on g.
These were tests of inductive reasoning,
visualization (the ability to imagine move¬
ments of relatively complex forms), quanti¬
tative reasoning, and verbal ability. He con¬
cluded that
The eventual interpretation [of gj must
resort to analysis of what processes are com¬
mon to the tasks used in the measurement
[of the abilities just listed] and to the anal¬
ysis of what attributes of such tasks are
associated with their difficulties.
Carroll, 1993, p. 597
23 Gignac, 2006.
24 Gustafsson, 1984.
25 Johnson et al., 2007. 1 calculated the loadings based
on the data presented on p. 103 of their report.
26 Carroll, 1993, p. 597.
96
HUMAN INTELLIGENCE
It is certainly unclear what single cogni¬
tive process underlies all four of the tests.
Accordingly, let us look at the correlated
traits hypothesis, the idea that there are
some cognitive functions which, although
not what we would normally think of as
intelligence in themselves, are required for
a large variety of cognitive tasks. There¬
fore, individual differences in these func¬
tions would produce g as a statistical phe¬
nomenon, even though no such thing as
g exists as an underlying quality of the
mind.
How would this work? Consider an anal¬
ogy to carpentry. Carpenters use the same
tools to make a great many things, ranging
from tables and kitchen cabinets to fences.
Suppose that there are individual differences
in the quality of tools available to different
carpenters. Instead of thinking of a mod¬
ern carpenter, imagine a bit of time trav¬
eling, where we ask a modern carpenter, a
medieval carpenter, a Bronze Age carpenter,
and a Stone Age carpenter to make us some
furniture, each using the tools appropriate
to the historic era. We would not think of a
saw, hammer, or adze as indicating carpen¬
try skill, in itself. Nevertheless, across the
ages, we could extract positive manifold in
making furniture solely because of the qual¬
ity of the tools.
Unlike carpentry tools (but like carpentry
skills], psychological processes are dynamic.
In an absolute sense, cognitive competence
rises through adolescence, and declines in
old age. Short-term memory processes and
the speed of making simple decisions show
a similar rise and fall with age. The same
principle applies to knowledge. Knowledge
acquired in mathematics can be utilized
while studying physics. One way to find
support for the correlated traits hypothe¬
sis is to show that practice in solving one
task involving general intelligence will pro¬
duce improvement on another, seemingly
very different task.
There are not as many such studies as
I would like to be able to cite. However,
two studies do nicely illustrate the issue.
In one, German children were taught to
verbalize their strategies when attacking
matrix problems, and then to solve the task
by verbal reasoning. This improved scores
on the (allegedly] nonverbal test.27 A sec¬
ond, and even more dramatic, test involved
training children to use the abacus.28 As
anyone who has learned to utilize the aba¬
cus can testify, this requires concentration
of attention and recognition of visual pat¬
terns. Sudanese children were given an age-
appropriate RPM test, and then divided into
experimental groups that received substan¬
tial training in the use of the abacus, and
control groups that did not. Prior to train¬
ing, both groups had equivalent RPM scores.
Following training, the experimental group
outperformed the control group.
These two very different studies can be
used to make the same point. There are
certain cognitive skills that are useful in a
variety of contexts. These include verbaliza¬
tion, which makes different aspects of the
problem open to conscious inspection, and
concentration of attention, which is a more
primitive operation. No one of these skills,
alone, is “general intelligence/’ but collec¬
tively they are. If the possession of one of
these skills is statistically associated with the
possession of others, positive manifold, and
its correlate, g, will result even though no
single cognitive skill can be pointed to as
“intelligence.”
Why should the skills that contribute to
general intelligence be correlated? One pos¬
sibility is that all these skills draw on a com¬
mon biological capacity, such as efficiency
of neural processing, and that there are sub¬
stantial individual differences in that capac¬
ity. This argument makes positive mani¬
fold, and g, manifestations of a biological
phenomenon.
Another possibility is that positive man¬
ifold emerges from interactions in which
the development of one cognitive skill facil¬
itates another. This is called mutualism.
For instance, practicing verbalization during
problem solving might facilitate the ability
27 Carlson & Weidl, 1992.
28 Irwing et al., 2008.
PSYCHOMETRIC THEORIES
97
to control attention. If this is the case,
the possession of key cognitive skills would
become correlated, over time, even though
they were initially uncorrelated.29
To summarize, the statistical evidence for
g, positive manifold, could be produced by
a pervasive general intelligence factor, or it
could be produced by correlations between
more specialized abilities. This leaves open
the question of what these abilities are. Sub¬
sequently in this chapter we will look at
a psychometric model that derives g from
more specialized, correlated traits. First,
though, we want to consider two reserva¬
tions about the generality of findings on gen¬
eral intelligence.
4.3.4. Reservations about g
The extent to which a given study sup¬
ports the g model depends on the extent to
which the tests used show positive manifold.
If the tests are highly correlated, the g model
is supported; if they are not, it is not. It
cannot be stressed too strongly that many,
many well-designed studies provide results
that are consistent with the g model. Never¬
theless, two reservations are in order.
The first is that psychologists who have
studied intelligence have been rather con¬
servative in their decision about what sort
of test is a test of cognition. The bulk of
the evidence for g has been obtained by
analyzing tests that fit into the “Drop in
from the Sky” paradigm. If, as I urged in
the first chapter, we expand our definition
of intelligence to include problem solving
in other situations, then other factors might
be found, and g might or might not drop
in power. What might happen is not what
would happen. Whether studies of cogni¬
tion in expanded situations would provide
evidence for g should be decided by empiri¬
cal research, not by the intuitions of people
who support or oppose the model] There is
a disturbing lack of data on this topic.
The second reservation is that studies
of specialized populations often do not
29 Van Der Maas et al., 2006.
show strong evidence for g. In part this is
simply a statistical phenomenon. Take the
case of college students at a highly selective
institution, such as Stanford, Harvard,
or Cambridge. These students have been
selected by a process that, to a consider¬
able extent, evaluates their general reason¬
ing ability. Therefore, within the selected
group there will be a restricted range of
g, and other factors will determine individ¬
ual differences in test scores. The situation
is analogous to the fact that height is not
closely related to the ability to score points
in professional basketball - because almost
all the players are already very tall.
This poses a practical problem, because
in general the populations that are easiest to
study are the ones where such restrictions
of the range of reasoning ability may occur.
We do not have to go to highly selective col¬
leges to see this effect; all college and uni¬
versity students have been subject to some
selection on general reasoning ability. The
same situation occurs when people try to
obtain voluntary samples from the general
population. A truly random sample is very
hard to construct. It is much easier to recruit
people from the middle and upper socioeco¬
nomic classes (SES] than to recruit people
with low SES. This fact, and many other
recruitment biases, operate to produce sam¬
ples with restricted ranges on g, thus often
underestimating the importance of the trait
in the population.
However, that is not quite all there is to
studies of populations with restricted ranges
of g. There are some systematic changes in
the ubiquity of g that are not solely due to
statistical phenomena.
4.3.5. Where g Is Not Found: There Are
Very Few Universal Geniuses
In the 1930s Leon Thurstone, a professor
at the University of Chicago, challenged
Spearman’s g model.50 Thurstone believed
that intelligence is based on several dis¬
tinct primary abilities, rather than on a single
30 Thurstone, 1938: Thurstone & Thurstone, 1941.
98
HUMAN INTELLIGENCE
general-reasoning factor. The primary abili¬
ties were:
Spatial reasoning : The ability to rea¬
son about figural representations. An
example would be deciding whether
two pictures did or did not represent
the same object viewed from different
perspectives.
Perceptual speed : The ability to detect
simple figures in a display. For an illus¬
tration, try to find the browser icon on
your computer desktop.
Number facility. The ability to do rela¬
tively simple numerical computations
quickly.
Verbal relations : The ability to compre¬
hend verbal statements.
Word fluency : The ability to produce
simple words and statements rapidly.
Memory. The ability to recall informa¬
tion. Thurstone did not distinguish
between short-term and long-term
memory, as we certainly would do
today.
Inductive reasoning: Spearman’s ability
to see patterns. A deductive reason¬
ing factor was added in some of
Thurstone’s work.
Thurstone claimed that these abilities are
essentially statistically independent; being
good or poor on one of them does not pre¬
dict whether a person is good, poor, or aver¬
age on another. This conclusion is diamet¬
rically opposed to Spearman’s claim that
intelligence is largely produced by a single
general-reasoning factor.
In the 1940s and 1950s there was debate
over whether the discrepancy between
Thurstone’s and Spearman’s results might
have been due to different groups having
used different factor analytic methods. The
development of modern computerized tech¬
niques has essentially ended that discussion.
The discrepancy was probably due in part
to restriction of range effects. In general,
Spearman and other British psychologists
analyzed data from the testing of schoolchil¬
dren. While Thurstone did similar studies,
he also relied a great deal on studies of
University of Chicago students. Chicago
was, and is, a highly selective institution, so
his college sample undoubtedly had a highly
restricted range of scores on g. A third likely
reason for the discrepancy has more psycho¬
logical content. It could be that the structure
of intelligence is more differentiated at high
levels than at lower levels. In concrete terms,
unusually high scores on a test of verbal rea¬
soning might be only moderate predictors of
unusually high scores on a test of mathemat¬
ical reasoning, while unusually low scores on
the verbal test could be very good predictors
of unusually low scores on the mathemati¬
cal test. To the extent that this is true, factor
analysis would reveal relatively independent
factors in a high-ability group, while reveal¬
ing a strong g factor in a low- ability group.
Modern research has shown that this
is the case. Douglas Detterman, a pro¬
fessor at Case-Western Reserve University,
divided the WAIS-R and WISC standard¬
ization samples into five ability groups.51
The strength of the g factor was highest
in the low-ability group, and declined as
group IQ score increased. Similar results
have been obtained by other investigators,
using other tests, in both national and inter¬
national settings.32
This is an important result, because it is
relevant to a social issue, the distribution of
intelligence at high levels of cognitive func¬
tioning. High levels of talent appear to be
fairly specialized, while marginal cognitive
talent seems to have general effects. Why
might this be the case?
In part it is probably due to experience.
Modern society encourages specialization to
a much greater extent than past societies
did. Studies of expertise in a variety of fields,
ranging from athletics to chess, have shown
that acquiring a high level of expertise takes
a great deal of time and effort.33 At high lev¬
els of talent, therefore, social pressures lead
to a differentiation of cognitive competences
31 Detterman, 1991; Detterman & Daniel, 1989.
32 Abad et al., 2003; Colom et al., 2002; Deary et al.,
1996; Hunt, 1995a; Legree, Pifer, & Grafton, 1996.
For some negative evidence see Sakolofske et al.,
2008.
33 Ericsson, 1996; Ericsson et al., 2006.
PSYCHOMETRIC THEORIES
99
due to specialized training. But this cannot
be the whole picture, because the differen¬
tiation of abilities at high levels (or, con¬
versely, generalization at low levels) occurs
in children, as evidenced by studies on dif¬
ferentiation involving the WISC.
Whatever the reason for this, we will
want to keep in mind the fact that high lev¬
els of talent are specialized when we come
to a discussion of the implications of intelli¬
gence, in Chapter 10.
4.4. The Three-Stratum Model: Cattell,
Horn, and Carroll's View of
Intelligence
A widely cited alternative to the g model
is the Gc-Gf or three-stratum model, orig¬
inally developed by Raymond Cattell and
John Horn, then modified by John Carroll.54
Cattell studied with Spearman in Eng¬
land, and then moved to the University
of Illinois and finally to the University of
Hawaii. He and his then-student at Illinois,
John Horn (subsequently a professor at the
Universities of Denver and Southern Cali¬
fornia), believed that Spearman’s theory did
not give sufficient weight to group factors.
They were also skeptical of the idea that
g is a trait. Instead they believed that posi¬
tive manifold is due to individual tests draw¬
ing on several broad factors. To follow their
argument, consider that bane of K-12 stu¬
dents, the mathematical word problem.
A train leaves station A and proceeds to
station B , traveling at 60 miles an hour.
At the same time that this train leaves A,
another train leaves B, bound for A, travel¬
ing at 30 miles an hour. The distance from
A to B is 2jo miles. How long will it be
before the trains meet?
A student who tries to solve this problem
has to know the meaning of words and the
syntax of English, lexical retrieval ability and
sentence comprehension. Several facts have to
be kept in mind as others are received, call¬
ing upon short-term memory ability. Numer-
34 Carroll, 1993; Cattell, 1971, 1987; Horn, 1985; Horn &
Noll, 1994.
ical facility is required. These narrowly
defined abilities are examples of first-stratum
or primary abilities. The first-stratum abil¬
ities are themselves grouped into broader,
second-stratum abilities. According to Cattell
and Horn the most important of these are
fluid intelligence (Gf) and crystallized intelli¬
gence (Gc), which they defined as the abil¬
ity to deal with new and unusual prob¬
lems (Gf) and the ability to apply previously
acquired knowledge to the current problem
(Gc) . In many contexts visual- spatial ability
(Gv), the ability to deal mentally with visual
images, is also important. The primary abil¬
ities of inductive and deductive reasoning
are grouped under fluid intelligence; general
cultural knowledge and lexical knowledge
abilities are grouped under crystallized intel¬
ligence; and the abilities to compare spatial
forms and to manipulate spatial forms “in
the mind’s eye” are grouped under visual-
spatial ability.
As the theory has evolved additional
second-stratum abilities have been defined.
These include factors for retrieval from
short- and long-term memory, the ability to
deal with auditory as well as visual stimuli,
quantitative ability, and two factors reflect¬
ing cognitive processing speed; cognition in
general [cognitive processing speed, Gs) and
another dealing with the speed with which
very simple decisions are made [decision
reaction time, Gt) .
Cattell and Horn had long and active
careers, during which they had colleagues
who conducted research dealing with such
things as the processing of auditory and tac¬
tile stimuli. In the typical extension of the
theory a study would be conducted with
a battery that included some tests already
identified as a marker of abilities found
by previous research, and some new tests
that explored different primary skills. This
process inevitably resulted in the defini¬
tion of still more second-stratum factors.
For instance, Stankov used this paradigm to
extend the Gf-Gc model to tasks involving
auditory stimuli, such as the ability to dis¬
criminate sounds.55
35 Stankov & Horn, 1980; Horn & Stankov, 1982.
lOO
HUMAN INTELLIGENCE
Figure 4.8. The structure of the second-stratum and third-stratum
abilities in the WJ-III test. Codes: Gc - crystallized intelligence;
Gf - fluid intelligence; Gq - quantitative ability; Gv -
visual-spatial ability; Glr - long-term storage and retrieval in
memory; Gsm - short-term memory; Gs - cognitive speediness;
g— general intelligence. The loadings are those found in the WJ-III
normalization sample, ages twenty to thirty-nine.
4.4.1. Extensions and Applications
of the Three-Stratum Model
On the basis of his substantial review of
the literature, Carroll36 concluded that the
Gc-Gf model provided a good fit to most
of the over 450 data sets that he reana¬
lyzed. However, in almost all cases abilities
at the second stratum were themselves cor¬
related. He took this as evidence for a single
third-stratum factor, general intelligence.
Conceptually this is the g factor advocated
by Spearman37 and, almost a century later,
by Jensen. 38
The resulting three-stratum theory has
been used as a basis for several psychometric
batteries, including the Woodcock-Johnson
test battery39 [WJ-III] and the revised
36 Carroll, 1993.
37 Spearman, 1904, 1927.
38 Jensen, 1998.
39 McGrew, 2005.
Kaufmann Adult Intelligence Test
[KAIT].40 Table 4.3 lists some of the
primary abilities evaluated by the WJ-III
battery. In the interests of space, only
enough of these abilities are listed to give
the reader an idea of the theory.41 In the WJ-
III most of these abilities are represented by
a single test, and therefore the structure at
the first-stratum (primary- ability] level, per
se, cannot be tested.
The heart of the theory lies in the
broader, second-stratum abilities. These are
shown in the left-hand column of Table 4.3.
Reasoning abilities fall under Gf, while
verbal comprehension and general knowl¬
edge abilities fall under Gc. The second-
stratum abilities are themselves correlated.
This gives the overall test the structure
shown in Figure 4.8. All the second-stratum
40 Alfonso, Flanagan, & Radwan, 2005.
41 McGrew, 2005.
PSYCHOMETRIC THEORIES
101
Table 4.3. A three-stratum grouping of primary abilities (middle and right-hand columns)
into secondary abilities (left- and right-hand columns) in the three-stratum theory. The
table is not complete, although the major secondary abilities are shown.
Second Stratum Subsumed Primary
Ability Abilities Brief Description
Fluid intelligence (Gf)
Crystallized intelligence
(Gc)
Deductive reasoning
Inductive reasoning
Quantitative reasoning
Linguistic development
Lexical knowledge
Information about
culture
The ability to solve novel problems.
The ability to reason from established
principles or facts.
The ability to detect patterns in
observations.
The ability to reason using numerical and
mathematical arguments.
The ability to apply previously acquired
knowledge to current problems.
The ability to follow an argument in one’s
native language.
The extent of one’s native-language
vocabulary.
Knowledge of the facts about one’s own
culture.
General domain-specific
knowledge (Gkn)
Geographic knowledge
Mechanical knowledge
Other tests of
knowledge within
various domains
Visual-spatial ability (Gv)
Visualization
Spatial relations
Closure speed
Short-term memory
[Gsm]
Memory span
Working memory
Breadth of knowledge of facts and
principles in specific domains.
The ability to deal with structured visual
stimuli.
Ability to recognize and match visual
stimuli.
Ability to manipulate visual stimuli “in the
mind’s eye.”
The speed at which familiar visual stimuli
can be recognized when obscured or
hidden in other stimuli.
The ability to apprehend and store
information about the current situation.
The ability to repeat back, verbatim,
information just presented.
The ability to execute cognitive processes
on information held in short-term
memory and to store the results.
(continued)
102
HUMAN INTELLIGENCE
Table 4.3 (continued)
Second Stratum Subsumed Primary
Ability Abilities Brief Description
Long-term storage
and retrieval (Glr]
Associative memory
Meaningful memory
Free recall
Cognitive processing
speed (Gs)
Pattern recognition
Numerical facility
Decision speed (Gt)
The ability to store information for long
periods of time and to retrieve it.
The ability to recall a complex piece of
information given a previously learned
association to it.
The ability to store, retain, and recall
meaningful pieces of information,
especially biographical information.
The ability to recall, without cues, a long
list of previously presented pieces of
information, where each piece is unrelated
to the others.
The ability to execute easy, highly
over learned cognitive tasks.
Quick recognition of familiar perceptual
patterns.
The ability to execute simple arithmetic
operations quickly.
Speed of responding in simple decision
situations.
Choice reaction time Choosing which of a small number of
pre-defined signals has been presented.
Semantic processing speed Deciding whether a string of letters is or is
not a word, e.g., CAT vs. TAC.
Mental comparison speed Time required to compare two familiar
stimuli on some attribute, e.g., deciding
whether the symbols 'A' and 'a’ name the
same letter.
Note: This table is based on a larger table presented by McGrew (2005), Table 8.3.
abilities have loadings on the single third-
stratum factor; g.
The WJ-III was standardized on a sample
of just under 9,000 examinees, of varying
ages, who were chosen to approximate the
distribution of a number of demographic
variables (sex, age, ethnicity] in the United
States. Table 4.4 presents the g loadings for
several of the second-stratum tests, sepa¬
rately for different age groups.42 The load¬
ings are both high (in the .80 range and
above] and remarkably stable over adult¬
hood. The very high loadings of Gc and Gf
42 McGrew & Woodcock, 2001, Appendix F.
on the g factor imply a correlation between
Gc and Gf in the .80-85 range.
The high loadings of second-stratum abil¬
ities on g have led some investigators to
argue that the Gf-Gc distinction is a need¬
less complication. For instance, Gustafsson
has reported a study of over 1,000 Swedish
sixth-graders in which the data was fit by
a three-stratum model, with three broad-
level abilities, Gf, Gc, and Gv. Gf had a
loading of 1 (identity!] on the third-stratum
g factor. The Gv and Gc loadings were
.80 and .76, respectively.4* Therefore, the
43 Gustaffson, 1984.
PSYCHOMETRIC THEORIES
103
Table 4.4. The g loadings of the second-stratum factors on the WJ-III test, omitting Ga
(auditory ability). Primary abilities are as shown in Table 4.3.
Broad Ability
Age 6-8
Age 9-13
Age 14-19
Age 20-39
Age 40+
Gc
■79
.86
■9°
■91
.92
Gf
.96
.89
•92
.92
•94
Gq
•73
•74
•7i
.66
.85
Gv
.85
.68
•77
•79
.85
Glr
.80
•75
.80
•95
.89
Gsm
.96
•91
.83
.86
.92
Gs
.61
•49
■49
■51
•75
Source: McGrew & Woodcock, 2001, Appendix F.
expected correlation between Gf and Gc
would simply be the Gc loading, .80. This
result contrasts with an earlier study, using
rather different factor analytic techniques,
in which correlations as low as .17 between
Gf and Gc were reported.44 I am somewhat
at a loss to explain this difference. Taking
the literature as a whole, it does appear
that in the majority of cases fluid and crys¬
tallized intelligence are substantially corre¬
lated. But does this mean that they are both
expressions of an overarching general intel¬
ligence, g?
4.4.2. The Nature of g, Gc, and Gf
in the Three-Stratum Model
In the three-stratum model the second-
stratum factors fall into three groups: the
two cognitive factors, Gc and Gf; the short-
and long-term memory factors; and factors
relating to different sensory modalities,
notably Gv (visual) and Ga (auditory) .
Almost any test involving cognition will
involve Gc and Gf, to some degree. How¬
ever, a battery might emphasize one or
the other type of cognition. Whether or
not memory or sensory modality factors are
found depends upon the particular batteries
used.
Two studies illustrate this point. Gustafs-
son’s previously cited study of Swedish mid¬
dle school children utilized a battery of thir¬
teen tests. He recovered three second-order
factors - Gc, Gf, and Gv. Only two of his
tests involved auditory presentations, and
they were both memory span tests. There¬
fore, a factor involving these two tests, only,
could be referred to as either a short-term
memory or auditory skill factor. Gustafsson
understandably chose to interpret variation
in performance as due to short-term mem¬
ory ability, as the only auditory component
of the test involved hearing and remem¬
bering words or numbers. The composition
of Gustafsson’s battery contrasts with the
extensive auditory tests (e.g., pitch discrim¬
ination) used in investigations that have suc¬
cessfully searched for Ga.45 These differ¬
ences certainly do not invalidate or weaken
Gustafsson’s study. They do drive home the
point that how many human abilities are
found depends upon how broad a range of
behaviors is studied.
The second example involves a case that
illustrates a common practice. Many inves¬
tigators have extracted a general factor
from tests that were designed for practical
rather than research applications, such as the
ASVAB and the SAT. Are the general fac¬
tors derived from these tests reflections of
the cognitive trait that is revealed by anal¬
yses of battery-type intelligence tests, such
as the WAIS and WJ? According to the g
model, they should be. Jensen, in particu¬
lar, has argued that the identification of g is
almost indifferent to the indicators used in
a battery 46
45 Horn & Stankov, 1982; McGrew & Woodcock, 2001.
44 Hakstian & Cattell, 1978. 46 Jensen, 1998, p. 91; 2002.
HUMAN INTELLIGENCE
104
From the viewpoint of the three-stratum
model this is simply not accurate, as the fol¬
lowing example shows.
The ASVAB was designed for a spe¬
cific purpose, predicting performance as an
enlisted person in the US military services.
Some of the tests used in the ASVAB are
combined into a general index, the AFQT.
[See Chapter 2 for a more complete descrip¬
tion.) A general factor that is highly corre¬
lated with the AFQT can be extracted from
the ASVAB.47 But is it g?
In their widely publicized Bell Curved
Richard Herrnstein and Charles Murray cor¬
related the AFQT with a variety of indica¬
tors of social behavior, ranging literally from
education to divorce rate to salaries, and
offered their results as evidence that gen¬
eral intelligence has a great deal of influence
upon success in our society. We will discuss
these results in some detail in Chapters 10
and 11. Herrnstein and Murray, and Jensen,
assumed that g, as revealed by the AFQT,
is equivalent to a general intelligence factor
uncovered by other test batteries. Extending
Gustafsson’s argument, this could be called
either g or Gf.
Richard Roberts and his colleagues at
the Educational Testing Service tested this
assumption. They combined scores on the
ASVAB with scores on tests designed to
measure Gf and Gc separately, and found
that the general factor on the ASVAB is a
measure of Gc rather than Gf.49 This is not
a small point. Virtually everyone acknowl¬
edges that Gc is responsive to education,
while Gf is less responsive. Regarding the
AFQT as a measure of Gc rather than a mea¬
sure of g or Gf increases our optimism about
the prospects of improving society through
education.
Studies such as Roberts and colleagues’
bring home the importance of a method¬
ological point made in section 4.2. A fac¬
tor is not an invariant property of a test. A
factor is a statistic produced by the inter¬
action between the abilities evaluated in a
47 Ree & Earles, 1991.
48 Herrnstein and Murray, 1994.
49 Roberts et al., 2000.
test battery and the distribution of those
abilities in the sample being tested. The
fact that g, Gf, and Gc keep reappearing
shows that these traits are somewhat invari¬
ant across different situations. That is infor¬
mative, for it suggests that they are indeed
basic dimensions of variation in human cog¬
nition. Blithely accepting the ubiquity of g
can produce misleading generalizations.
4.4.3. What Is a Natural Kind; g, Gf
or Gc?
A natural kind is a phenomenon that exists
in nature and is to be discovered. This
contrasts with an artifactual classification,
which is constructed by human thought.
The distinction between men and women is
a natural kind; the distinction between legal
and illegal residents of the US or the Euro¬
pean Union is an artifactual one. Artifactual
classifications can be quite useful in some
settings. Nevertheless, they are the results
of human action, not the action of a law of
nature.
In studies of intelligence the broad sen¬
sory modality and memory factors are natu¬
ral kinds, for they are defined as individual
differences in human biological capacities.
The distinctions among Gc, Gf, and g are
debatable.
The Gf-Gc distinction is based upon
an individual’s social and cultural history.
Consider how the following problem might
be solved:
Two freight trains approach each other on
a single track. Between them there is a
brief side route capable of holding only one
engine or box car. (See Figure 4.9.J How
can the two trains pass each other?
Most readers will see this as a novel prob¬
lem, and solve it using a mixture of “reason¬
ing about new problems” [Gf) and manipu¬
lation of visual images [Gv). In fact, it is a
problem that occurs often in railroad switch¬
ing yards, and has a standard solution. 5° The
railroad-experienced reader will treat the
problem as an exercise in Gc.
50 Hayes, 2007.
PSYCHOMETRIC THEORIES
105
uiiimii i i i^i^xLLmmiiiiiiiiiiiiiiiiiiiimTrnT^j ri m i m m i m itttttth
• Engine
I I Boxcar
Figure 4.9. The railroad passing problem. Two trains approach each
other. There is a side route that can hold only one engine or boxcar.
Is it possible for the trains to pass each other and, if so, how?
Test developers get around such issues
by restricting tests of Gc to tasks that draw
upon “generally culturally accepted knowl¬
edge.” This automatically restricts most Gc
tests to the industrially developed coun¬
tries, although the concept of “cultural
knowledge” applies equally well to, say, a
nomadic or hunter-gatherer culture. How¬
ever, it would have to be tested using ques¬
tions appropriate to the hunter-gatherer cul¬
ture. A test of Gc in one culture could be a
test of Gf in another. This is an important
distinction, for the two types of intelligence
make different demands on our information¬
processing capacities.
In order to solve a novel problem you
must develop a way of representing it men¬
tally. This can be a difficult task, involving
the development of information structures
to be held in short-term (“working”) mem¬
ory. In order to solve a problem by apply¬
ing previously acquired knowledge you must
have had appropriate experiences and coded
them in long-term memory in such a way
that they are accessible in the present con¬
text. Working and long-term memory draw
on different brain processes and structures.
This argument suggests that Gc and Gf are
two different natural kinds of ability.
This argument is not vitiated by demon¬
strations that g exists as a statistical phe¬
nomenon. Cattell argued that we acquire Gc
very largely by using Gf to discover appro¬
priate problem-solving techniques, a process
that he referred to as investing Gf in the
acquisition of Gc. If two people enter into
some experience that is unfamiliar to both
of them, the one with the higher Gf will
learn more from the experience, and end
up with a higher Gc than the other per¬
son. Repeated throughout life, this would
produce a correlation between Gf and Gc,
and hence statistical evidence for g even
though there is no natural kind that corre¬
sponds to general intelligence.
The contrary argument, made most
notably by Jensen, is that there are pervasive
individual differences in brain processes that
cause general competence, and that these
differences are causes of differences on both
Gc and Gf tasks.51 There is no way to decide
between the Cattell and Jensen positions
from an analysis of the psychometric data
alone. There is an important nonpsychome¬
tric argument in favor of Cattell’s proposal.
Experimental psychologists show that
two underlying human capacities are dif¬
ferent by showing that they respond to a
change of conditions in different ways. For
instance, one of the strongest pieces of evi¬
dence for a distinction between conscious
and unconscious memory systems is that
conscious memory retrieval falters when
a person is distracted, while unconscious
retrieval does not.52
This sort of argument can be used to dis¬
tinguish between Gc and Gf, using a uni¬
versally occurring experimental condition,
aging. Performance on tests that mark for
Gf (e.g., matrix problems) declines with
age over the adult life span. Performance
on tests that mark for Gc (vocabulary tests
and tests of cultural knowledge) does not.
In fact, measures of Gc may increase until
advanced old age.53 This provides striking
51 Jensen, 2006.
52 Jacoby, Toth, & Yonelinas, 1993.
53 Horn, 1985, 1986. Horn argued that the decline of Gf
began as early as the late 1920s. Horn relied mainly
on data from cross-sectional studies. This may have
led him to overestimate the decline, because it has
since been shown that there are cohort effects on Gf
io6
HUMAN INTELLIGENCE
Figure 4.10. Vernon’s structure of intelligence
model. The model contains a g factor and
two orthogonal group factors - the
verbal -educational factor and the
perceptual-mechanical skill factor.
evidence that Gf and Gc are indeed two sep¬
arate processes. We look at this evidence in
more detail in Chapter 11.
The body of evidence favors the three-
stratum theory over a simple general intelli¬
gence model. However, a revision of g the¬
ory ; the g-VPR model, deals with the evi¬
dence even better.
4.5. Johnson and Bouchard's
g-VPR Model
By 2000 the three-stratum model had been
widely accepted. Then, in 2005, two Univer¬
sity of Minnesota scientists, Wendy John¬
son and Thomas Bouchard, Jr., published
new analyses that both questioned the three-
stratum model and offered an alternative.54
Johnson and Bouchard began with a
model of intelligence that had been pro¬
posed by the Canadian psychometrician
Philip E. Vernon in the 1960s,55 but then
almost forgotten. Vernon himself barely
mentioned his model in a 1979 book on
heredity, environment, and intelligence!56
marker tasks. These are discussed in more detail in
Chapters 10 and 11. For the point being made here it
does not matter whether the cause of the change in
Gf and Gc scores is due to aging per se or due to
cohort effects. The important finding is that the two
abilities change differently over time.
54 Johnson & Bouchard, 2005a.
55 Vernon, 1964, 1965.
56 Vernon, 1979.
Vernon’s structure of intelligence model is
shown in Figure 4.10. It contained three fac¬
tors: a general factor (g) and two factors
orthogonal to g. One he identified as a ver¬
bal: educational factor, reflecting the empha¬
sis upon verbal skills in the educational sys¬
tem, and the other as a perceptual: motor
factor representing skill in identifying and
manipulating objects. Vernon also suggested
the presence of a third special factor, math¬
ematical skills, but felt that this was related
to the perceptual: motor factor.
Johnson and Bouchard proposed a four-
stratum model, with the structure shown in
Figure 4.11. The first, bottom, level consists
of the primary traits evaluated by individual
tests, such as a test of the ability to do sim¬
ple computations or to solve anagrams. The
second stratum consists of broader but still
fairly narrow abilities. For instance, at this
level there is a distinction between word
fluency, which is essentially a measure of
speed of producing verbal associations, and
verbal comprehension (their term, "verbal”),
which is characterized by vocabulary and
the understanding of proverbs. A similar
distinction was made between memory for
meaningful material and memory for arbi¬
trary, experimenter-presented associations,
such as arbitrary lists of number-noun pairs.
In all the data sets analyzed they found sub¬
stantial correlations between second-level
factors, which indicated a need for a third
stratum, in which the number of factors
would be reduced, and where a second-level
factor could have loadings on more than one
third-order factor.
The third stratum, which is the heart of
the model, contains three factors - Vernon’s
verbal and perceptual skills factors and a
third "perceptual” ability, the ability to envi¬
sion motion of a static figure, most clearly
seen in tasks that require rotation of a visual
figure “in the mind’s eye.” As was the case for
Vernon’s model, the Johnson and Bouchard
model does not contain a memory factor.
This is consistent with research in cognitive
psychology, which has shown that there are
many different types of memory.
Johnson and Bouchard’s third-level fac¬
tors were highly correlated, indicating a
PSYCHOMETRIC THEORIES
107
Narrow verbal skills, e.g.,
verbal comprehension.
Figure 4.11. The structure of Johnson and Bouchard’s VPR model.
In the interests of clarity, the figure shows a hierarchy. However,
the model is better described as having a lattice structure, for
individual tests may have loadings on more than one
second-stratum factor, and second-stratum factors may have
loadings on more than one third-stratum factor.
need for a fourth stratum. They found
that they needed only one factor at this
level, which they identified as general intel¬
ligence, g.
Developing an acronym from the names
of the third-level factors, Johnson and
Bouchard refer to their model as the g-VPR
model.
4.5.1. Psychometric E vidence
for the g-VPR Model
Johnson, Bouchard, and their colleagues
have presented two arguments for prefer¬
ring the g-VPR model to either the g model
(which it amplifies, rather than replaces)
or the three-stratum Gf-Gc model. The
first argument is based on psychometric
evidence, the second on biological plausi¬
bility. We look first at the psychometric
evidence.
Johnson and her colleagues made a com¬
parative analysis of three different data sets,
in which they compared the g-VPR model
to the Gc-Gf model and Vernon’s model of
orthogonal verbal and spatial abilities.57 The
first data set came from 400 adults who par¬
ticipated in the Minnesota Study of Twins
Raised Apart. This is an extensive study
of twins and their relatives, which will be
described in detail in Chapter 8, panel 8.7.
For the present purposes, we can think of the
study as a large data base in which the same
adults completed three different battery-
type tests. The investigators then reanalyzed
one of the data sets Thurstone had used to
justify his primary mental abilities model.
Thurstone 's data set contained scores from
a study done in Chicago in the 1930s. Some
57 Johnson & Bouchard, 200 5 a, b; Johnson, te Nijen-
huis, & Bouchard, 2007.
io8
HUMAN INTELLIGENCE
of the participants had taken sixty different
tests. The third data set was based on 46 dif¬
ferent tests given to just over 500 seamen in
the Dutch Navy in the 1960s.
In all three data sets, the g-VPR model
fit better than any of the other models. A
comparison to the three-stratum model is
particularly informative. Johnson and her
colleagues subsumed Gc into a verbal fac¬
tor (V), identified the Gf factor with g, and
split the Gv factor into their two perceptual
factors, roughly the analysis of static visual
images (P] and the ability to conduct mental
manipulations of visual objects (R). Because
none of the batteries that they considered
contained tests involving auditory presenta¬
tions, they had no opportunity to uncover
an auditory [Ga] factor.
The V, P, and R factors were not orthog¬
onal, but, as would be expected, the cor¬
relation between the P and R factors was
higher than the correlation between the V
factor and the other two. While none of
these samples can be claimed to be rep¬
resentative of a particular population, as
opposed to the standardization samples for
the WAIS, ASVAB, and WJ tests, one has
to be impressed by the uniformity of the
results, obtained over a wide variety of tests
and using markedly different samples.
Johnson and Bouchard did not find a need
to identify broad memory factors, although
they did identify some specific memory
factors in the second stratum. This does
not mean that memory is unimportant to
cognition; obviously it is. What it means is
either that memory factors are subsumed by
one of the broad reasoning factors or that
memory abilities are highly specific to the
type of material being memorized. As we
shall see in Chapter 6, there is evidence
for subsumption of the ability to deal with
short-term memories into g, and to view
long-term memory abilities as being highly
specific.
Johnson and Bouchard regard their psy¬
chometric evidence as a disconfirmation of
the three-stratum model. This is a strong
statement. Whether you accept it or not
depends upon your approach to statistical
hypothesis testing.
If one takes the classic approach to
hypothesis testing, none of the models,
including the g-VPR model, accounted for
the data. In every comparison there was a
“statistically significant” deviation of the data
from that predicted by the Gc-Gf, three-
stratum, original Vernon, or VPR model.
Johnson and Bouchard took the relativis¬
tic approach of comparing the models to
each other, using a sophisticated elabora¬
tion of Bayes' law.58 This analysis identifies
the best model within a set of models to
be compared, rather than testing to see if
there are significant deviations from a partic¬
ular model. The g-VPR model was the clear
winner in this competition, although it did
not account for all the data in an absolute
sense.
4.5.2. Logical Arguments
for the g-VPR Model
Johnson and Bouchard were also critical of
the Gf-Gc distinction on logical grounds. I
will present an expansion on their argument.
Gc is supposed to represent the use of
previously acquired knowledge to solve the
current problem. But how are you to con¬
struct a test for this ability? It makes little
sense to test a person's level of a skill unless
the examinee has had a chance to acquire
the skill. Therefore, tests of Gc have to be
tests based upon information that is widely
available in the examinee’s culture. Indeed,
most Gc markers are of this nature. The
vocabulary tests used as markers of Gc are
roughly at the level of vocabulary used in
television dramas.
Suppose that tests have been constructed
such that we can be certain that every exam¬
inee has been exposed to the information
needed to do well on the test. Individual
differences in test scores will then be pro¬
duced by differences in examinees' ability
to extract this information from their com¬
mon experiences. A great deal of cultural
knowledge is based upon induction from
58 See Hunt, 2007, for a discussion of Bayes' law. The
test used was the Bayesian Information Criterion
(BIC) developed by Raftery (1995).
PSYCHOMETRIC THEORIES
109
experience, rather than explicit instruction.
This is especially true of our ability to under¬
stand the meaning of words in different con¬
texts. "How are you feeling?" spoken by a
waitress at a restaurant is a different ques¬
tion, and requires a different response, than
“How are you feeling?” spoken by a physi¬
cian in an emergency room. In order to
understand such distinctions a person has to
recognize patterns of usage. Pattern recogni¬
tion is, by definition, Gf, and Spearman's g.
This muddles the Gc-Gf distinction. A
test of Gc is either not fair, because it evalu¬
ates information to which the individual has
not been exposed, or it is actually a disguised
test of Gf.
Johnson and Bouchard also point out
another problem with the Gc-Gf distinc¬
tion, one that had actually been raised by
Horn some time earlier.59 If Gf is an ini¬
tial ability that is invested to produce Gc,
then Gf measures should be more respon¬
sive to individual biological variables than
Gc measures. In at least one case this is
simply not what happens. The heritability
coefficients for Gc and Gf are approximately
the same, in violation of the argument that
Gc reflects cultural experiences, while Gf
does not. By contrast, heritability analy¬
ses (to be discussed in Chapter 8 in more
detail) show a coherent pattern of genetic
association for the variables in the g-VPR
model.60
The g-VPR model aligns with well-known
neuroscientific findings. The ubiquity of g
suggests that there are individual differ¬
ences in some pervasive brain processes.
Three candidate processes have been sug¬
gested: individual differences in the ability
to control attention, individual differences
in the speed and accuracy of neural con¬
duction, and individual differences in the
plasticity of neural connections, all of which
would affect the ability to acquire and
retrieve information. Language processing
and perceptual processing are carried out
by separate brain systems. At least one bio¬
logical distinction mirrors the distinction
59 Horn, 1998.
60 Johnson et al., 2007.
between mental rotation and the analysis of
static figures. There are large male-female
differences, in favor of males, in the ability
to conduct rotation-like tasks.61 The male-
female differences in other perceptual tasks
are much smaller, and in some cases are in
favor of women.
All in all, Johnson and Bouchard make a
persuasive case for their model.
4.6. Summary and Evaluation
of Psychometric Theories
Which psychometric theory is correct?
Any psychometric theory of intelligence
has to account for the degree to which posi¬
tive manifold exists. People who do well (or
poorly) on a test of one type of cognitive
skill generally do well (or poorly) on tests
of other cognitive skills. This tendency is
stronger at the bottom than at the top; poor
performance on test A is a better predictor
of poor performance on test B than good
performance on A is of good performance
on B. There is very little, if any, evidence
for a negative correlation between cogni¬
tive skills. People who are highly competent
with language may not be as good at visual
or mathematical tasks as they are at verbal
tasks, but they are unlikely to be bad at those
tasks. The opposite reasoning holds. Math¬
ematicians and physical scientists are gener¬
ally not novelists, although there have been
a few exceptions, but they are very unlikely
to be functionally illiterate.
These facts indicate that a theory of intel¬
ligence has to include something like g.
But g, alone, is not enough. The debate is
over the appropriate structure of the broad,
but not completely general, abilities that lie
below g in the structure of intelligence.
The g, three-stratum, and g-VPR mod¬
els can all account for the psychometric
data in a general way. If we look at rela¬
tive accuracy, the extensive data gathered by
Johnson and her colleagues strongly indicate
that the g-VPR model is a statistical winner.
However, statistical criteria are not the
61 Halpern, 2000.
no
HUMAN INTELLIGENCE
only criteria by which a theory should be
judged. Two other issues are relevant: the
utility of a theory as a decision aid and
the extent to which a theory at one level,
here a theory of psychometric data, fits into
theories and facts at a more basic level, here
information-processing and biological mod¬
els of thought.
In many personnel classification and
training situations the Gc-Gf model has a
great deal to recommend it. It makes sense
to distinguish between candidates who do
not know but can learn (low Gc, high Gf)
and those who already have the requisite
knowledge (high Gc]. It is also worth not¬
ing that in industrial and military personnel
screening the issue is usually whether a per¬
son has a particular cognitive capability, not
how he or she came to have it.
Both the Gc-Gf and g-VPR distinctions
can be fit into information-processing mod¬
els. The Gc-Gf model matches the distinc¬
tion between problem solving based on the
manipulation of working memory and prob¬
lem solving based on retrieval of previously
acquired information. The g-VPR model
matches the distinction between brain struc¬
tures involved in working memory and the
control of attention (i.e., the g and Gf com¬
ponents of the models can be identified with
the same brain structures]. The VPR com¬
ponents of the g-VPR model are closely tied
to pathways of sensory information process¬
ing. As Carroll noted, the Gc component
of the three-stratum model is very closely
associated with verbal reasoning.62
While g implies positive manifold, so do
the other models. Positive manifold could
be produced by the development of separate
modules of cognition that had positive influ¬
ences on each other’s development. Cattell
and Horn’s idea that Gf is invested in learn¬
ing to produce Gc, creating positive mani¬
fold as an epiphenomenon, is an example of
this sort of argument.
My conclusion is that a pure g model
is too simplistic, although useful in some
situations. Whether one should favor the
g-VPR or three-stratum model depends a
good bit on what the theory is supposed to
do - provide a starting place for a theory
that connects psychology to biology, or pro¬
vide a model for using human intelligence in
academia and the workplace.
Saying that the way you describe the sit¬
uation depends upon the problem you want
to solve is more in the spirit of engineer¬
ing than of pure science. Both are legitimate
worldviews.
And there is another worldview. All of
these models are attempts to find structure
in the data from conventional cognitive test¬
ing. In the next chapter we look at theories
that attempt to expand theories of cognition
beyond the data obtained from the “Drop in
from the Sky” paradigm.
62 Carroll, 1993.
CHAPTER 5
Taking Intelligence Beyond
Psychometrics
Newton said that if he had seen further it
was because he had stood on the shoulders
of those who had preceded him. In
Psychology we stand on their faces.
The late R. C. Bolles, Professor
and historian of Psychology, the
University of Washington, personal
communication
Walter Lippmann had more to say about
intelligence than was quoted in Chapter 1.
[ The intelligence test] does not weigh or
measure intelligence by any objective stan¬
dard. It simply arranges a group of people
in a series from best to worst by balanc¬
ing their capacity to do certain arbitrarily
selected puzzles, against the capacity of all
the others.
Lippmann, 1922b
This criticism strikes at the heart of psy¬
chometric theory. Lippmann believed that
“real” intellectual talent has little to do with
a talent for test taking. His criticisms have
been echoed, with surprisingly little change,
in today's world.
The gist of the modern criticisms of intel¬
ligence testing is that the tests capture a very
narrow slice of human cognition. The fol¬
lowing quotes are from books written for
the general public:
Almost everything you know about intelli¬
gence - the kind of intelligence psychologists
have most often written about - deals with
only a tiny and not very important part of
a much broader and more complex intel¬
lectual spectrum.
Sternberg, 1996, p. 11
The score on an intelligence test does pre¬
dict one's ability to handle school subjects,
though it foretells little of success in later
life.
Gardner, 1983, p. 3
The evidence does not warrant such skep¬
ticism. In fact, perhaps because of a per¬
ceived need to keep things simple in popular
books, these quotes are oversimplifications
of the authors’ opinions. Both Gardner and
Sternberg are considerably more circum¬
spect when they write papers addressed to a
professional audience.
111
112
HUMAN INTELLIGENCE
Nonetheless, the public clearly wants to
hear the attacks on testing. Why?
Possibly one of the strongest reasons is a
gut feeling that tests with at best a tenu¬
ous resemblance to real life problem solv¬
ing should not predict how well people can
solve important everyday problems. This
was the basis of Lippmann’s objection to
the first Stanford-Binet test battery. Imagine
what Lippmann would have to say about a
Progressive Matrices test]
Distrust is magnified by a certain amount
of defensiveness. Burt Green, a respected
psychometrician at Johns Hopkins Univer¬
sity, once told me that people think that a
fair question is one that they can answer.
There is something to this. Sternberg opens
one of his popular books with a diatribe
about how he, personally, did poorly on an
intelligence test.1 His anecdote will resonate
with anyone who feels that their test score
does not reflect their true mental capability.
Anecdotes about how tests failed to pre¬
dict outcomes are very common. There are
many stories of people who did well despite
their IQ, and a few stories of people (whom
the story teller usually did not know or
did not like) who did poorly even though
they had high IQs. I have never met anyone
who complained that their IQ score over¬
estimated their ability.
Those defending testing reply The plu¬
ral of “anecdote” is not “data” No one has
ever claimed that IQ scores are either a
perfect predictor or the only predictor of
social outcomes. Depending upon the sit¬
uation and the statistical assumptions one
wants to make, the correlations between
the trait underlying cognitive test scores and
either academic or industrial performance
are on the order of .5 or even higher. (See
Chapter 10 for elaboration.) Cognitive tests,
in some form, are given to literally mil¬
lions of people annually, so there is plenty
of room for anecdotes about how a test
did not predict for a particular individual,
even though the test is a valid predictor for
the population as a whole. The argument is
valid, but it will resonate more strongly with
1 Sternberg, 1996, p. 13.
members of the American Statistical Asso¬
ciation than with members of the general
public, for statistical reasoning is very much
an acquired skill.2
Another reason that people may dis¬
trust the tests is due to valid but limited
personal experiences. Interviews with “suc¬
cessful people,” including managers, reg¬
ularly produce assertions that personality
variables, such as self-discipline and open¬
ness to experience, are more important than
intelligence. This contradicts the evidence,
for in fact the correlations between indices
of success in the industrial world and per¬
sonality variables are about half the correla¬
tions between success and intelligence test
scores.3 Why the discrepancy?
People’s personal social circles are highly
selective. College graduates generally asso¬
ciate with other college graduates; industrial
workers talk with other industrial work¬
ers. Intelligence plays a role in determining
the people we know. Therefore the extent
of variation in intelligence among one's
acquaintances is likely to be smaller than
the variation in intelligence in the general
population. Providing that we avoid out¬
right psychopathology, there is no compa¬
rable restriction on variations in personality
traits. Intelligence may have less influence
over behavior than personality variables do,
within each person’s own social circle, but
more influence over behavior than personal¬
ity variables do, when considered across the
entire society.
Finally, two philosophic biases may lead
some people, especially social liberals (the
dominant social philosophy among aca¬
demic writers), to downplay the concept
of intelligence, as defined by psychometric
tests.
Test scores are correlated with famil¬
ial socioeconomic status (SES); the higher
your family’s SES, the higher your test
score is likely to be, and vice versa. There¬
fore, there is concern that using test scores
in personnel selection, and especially as a
2 Amsel, Langer, & Loutzenhiser, 1991; Gigrenzer
et al., 2007.
3 Hunt, 1995b.
TAKING INTELLIGENCE BEYOND PSYCHOMETRICS
criterion for admission to higher education,
will reinforce the existing social order. The
concern is somewhat ironic, because one of
the purposes of adopting cognitive tests as
screening devices was to reduce the hold
that the upper social class had on access
to higher education!4 Nevertheless, the con¬
cern is valid. If test scores alone were to
be used in admitting people to educational
institutions, there would be a tendency to
select applicants from families with moder¬
ate to high SES. The extent to which one
sees this as a problem depends upon one’s
philosophy about the purpose of higher edu¬
cation, a topic that is far beyond the scope
of a discussion of intelligence.
The second philosophic bias is closely
related to the first. There are differences
between demographic groups on virtually
all cognitive test scores, including the SAT,
ACT, and ASVAB. In the United States
Whites and Asian- Americans generally have
higher scores than African Americans and
Latinos. The gap in test scores mirrors the
gap in various measures of socioeconomic
status, such as income and health statistics.5
If you accept the test scores as valid indi¬
cators of intelligence, in the broad sense,
then you may appear to have accepted dif¬
ferences in intelligence as a partial explana¬
tion for the gap in the socioeconomic statis¬
tics. This important and very complex issue
is discussed in detail in Chapter 11.
Wanting something to be so, even for the
best of reasons, does not make it so. What
sort of evidence would make us want to
either replace or expand the model of intel¬
ligence developed from psychometric stud¬
ies? In order to answer that question we have
to consider how intelligence, in the concep¬
tual sense, rather than intelligence in the
sense of test scores, influences behavior. This
requires an amplification of the argument
presented earlier, in Chapter 1.
The problem is illustrated diagrammati-
cally in Figure 5.1. What we are ultimately
interested in is how intelligence produces
4 Lemann, 1999. See the discussion of the develop¬
ment of the SAT in Chapter 2.
5 Herrnstein & Murray, 1994.
u3
socially relevant behaviors. To investigate
this we have to measure both intelligence
and the behaviors. In the case of academics,
we might want to know how intelligence,
in the conceptual sense, determines what
a person learns in school. This is a rela¬
tion between concepts. All we can mea¬
sure is the relation between test scores and
grades, while admitting that the test scores
do not perfectly measure intelligence and
that grades do not perfectly measure what
students have learned. A similar argument
can be made for using test scores to evaluate
workplace performance.
Intelligence is not the only thing influ¬
encing grades, and for that matter it is not
the only thing influencing test scores. Grades
will be influenced by a variety of nonintel¬
lectual factors, such as self-discipline and
interest. Conventional psychometric test
scores can also be influenced by nonintellec¬
tual factors. Sternberg amplified upon his
own example by claiming that his IQ test
score was dramatically influenced, not for
the better, by a definable personality charac¬
teristic called test anxiety, which means what
the name implies, a tendency to become
panicked and underperform when tested.6
To further complicate things, social
behavior is not solely the product of the
individual doing the behaving. Behaviors can
be elicited or constrained by properties of
the situation. Gardner has provided a com¬
pelling example, in a series of biograph¬
ical essays on creativity, as illustrated by
such disparate figures as Einstein, Picasso,
T. S. Eliot, and Mahatma Gandhi.7 Each of
the creative geniuses Gardner wrote about
benefited from the support of people who,
often at considerable sacrifice to themselves,
played supporting roles so that the geniuses
could concentrate on the work that, ulti¬
mately, made them famous.
The following sections examine several
attempts to expand traditional psychomet¬
ric models of intelligence. I will start with
the simplest approach, expanding the con¬
ventional range of cognitive tests, and then
6 Sternberg, 1996, Chapter 1; Sarason, 1980.
7 Gardner, 1993b.
n4
HUMAN INTELLIGENCE
Figure 5.1. The relation between underlying traits, social constraints,
test scores, and socially relevant behaviors. Test scores can predict
but do not cause socially relevant behaviors. Both the behaviors and
test scores are constrained by noncognitive variables.
move to a discussion of personality and
motivational issues.
5.1. Gardner's Theory of Multiple
Intelligences
Howard Gardner's Multiple Intelligences
(MI] model, presented in his 1983 book
Frames of Mind? became instantly popular
as an alternative to psychometric theories of
intelligence. Frames of Mind was selected by
five book clubs and translated into seven lan¬
guages. The theory has since received a few
modifications, but the basic message has not
changed. 9 While Gardner has stressed the
application of his ideas in grade school edu¬
cation, he has not hesitated to discuss possi¬
ble applications in other domains.
8 Gardner, 1983.
9 Gardner, 1993a, 1999; Chen and Gardner, 2005.
As the name implies, MI theory is based
on the assumption that there are many
types of intelligence, ranging from linguis¬
tic skills to musical intelligence and bodily/
kinesthetic intelligence. Gardner regards the
purpose of schooling to be the development
of each child’s type of intelligence, without
forcing all children to focus on developing a
narrow set of linguistic and reasoning skills
that, he feels, are needed much more in con¬
ventional schools than they are in the world
outside school. While Gardner has made this
contention in many forums, to my knowl¬
edge he has never offered evidence showing
that the validity of IQ and similar educa¬
tional tests is confined to academics.
Why did Gardner’s view become so pop¬
ular? One reason is that he presents an opti¬
mistic view of education. MI theory offers
hope that the child who is not doing well
in conventional subjects may have supe¬
rior talents in art, music, social interactions,
TAKING INTELLIGENCE BEYOND PSYCHOMETRICS
or even bodily movement. It follows, then,
that it is the teacher's job to identify and
develop those talents. This approach to edu¬
cation is morale-building for students, par¬
ents, and teachers. The students and parents
are assured that there are ways to success
for every child. The teacher is encouraged,
for the theory stresses the teacher’s role as
a diagnostician and facilitator. This is a far
more professional view of teachers than the
view that the role of the teacher is to trans¬
mit the assigned lesson to the class, en masse.
By contrast, proponents of the general
intelligence (g) model are often interpreted
by educators as saying “some children are
smart and some aren’t and there is not much
that you, the teacher, can do about it.”
Therefore, the teacher’s role is to transmit
information, which either will or will not be
picked up by the students, depending upon
their pre-set intelligence.
In fact, g theorists have never said this.
What they have said is a somewhat subtler
message: “Some children are smart and some
are not. There is a great deal that you can
do to raise (or lower) the average level of
competence in a class, but there is little you
can do to lower the fact of variation. The
bright children will learn more than the not-
so-bright children, and this is a fact that a
teacher has to live with.”
The MI model is far more egalitarian than
the g model. According to it, if the cultural
situation (i.e., the classroom) is appropri¬
ately shaped, every child can have his or her
potentials drawn forth. Gardner tells teach¬
ers that they can make a difference for all
children, providing that they structure the
classroom so as to encourage the appropri¬
ate talent.
It would not be logically contrary to any¬
thing Gardner has written to assert that even
if there are many intelligences, there are
some children who simply do not have much
of any potential. However, that message is
downplayed, and to some extent negated, by
Gardner’s assertion that the different intel¬
ligences he has identified are not highly cor¬
related. This is an assertion about data, and
can be evaluated - as we shall do.
ll5
Now let us take a closer look at the
theory.
5.1.1. The Multiple Intelligences
MI theory is based on two premises. The
first is that there is no general trait of over¬
all mental competence. In Gardner’s view
positive manifold is a statistical artifact that
arises because conventional schools empha¬
size only a limited range of skills - language,
abstract reasoning, and rapid responding -
and because virtually all the tests are pre¬
sented using language. In Gardner’s view,
psychometric studies look at intelligence
through a “verbal lens.”
The second premise is that there are, in
fact, a variety of different intelligences, rang¬
ing from linguistic intelligence to bodily/
kinesthetic intelligence. But what are they?
The way Gardner identified the dimensions
of intelligence could not have been further
from the way the psychometricians investi¬
gate the topic. He says of his preparation for
his first book, in 198 3: 10
/ had always been intrigued by the chal¬
lenge and promise of examining human
cognition through a number of discrete
disciplinary lenses. I enjoyed investigating
psychology, neurology, biology, sociology,
and anthropology as well as the arts and
humanities. And so I began reading sys¬
tematically in these areas in order to gain
as much information as possible about the
nature of various kinds of human faculties
and the relationships among them.
Gardner , 1999, p. 33
This method of analysis, broad reason¬
ing followed by reflection, is quite differ¬
ent from the emphasis on data collection
and analysis that characterizes psychomet¬
ric research. Gardner himself has observed
that he has little concern for measurement.11
This makes his method of inquiry more akin
to inquiry in the humanities than to scien¬
tific inquiry. That does not make the method
10 Gardner, 1983.
11 Gardner, 2006a, b.
HUMAN INTELLIGENCE
116
wrong; humanistic inquiry is a legitimate
method of investigation. It has lead Gard¬
ner to an updated definition of intelligence:
A biopsychological potential that can be
activated in a cultural setting to solve prob¬
lems or create products that are of value to
society.
Gardner, 1999, pp. 33-34
This definition has been amplified by
the identification of four criteria for an
intelligence:12
Biological: The trait in question should
be the product of a biological system,
and therefore it should be isolatable
by brain damage and should exhibit
a development over evolutionary
history.
Psychological: The trait should be improv¬
able by specific training. Indicators of
the trait should be correlated with
other indicators of the same trait, but
should have only a low correlation
with indicators of other traits.
Developmental: The trait should have a
distinct developmental history, with
a definable set of expert-level end-
state performances. There should also
be special populations who display
unusual development of the trait, such
as idiot savants (who have a single
exceptional talent, but are otherwise
somewhat or even profoundly below
the norm] and prodigies.
Logical: Each intelligence should have a
definable core set of operations that
can be expressed in a symbol system.
The obvious case here is language,
which is a symbol system whose struc¬
ture and use are determined by the
rules of syntax, semantics, and prag¬
matics appropriate to the language and
culture involved. Music and chore¬
ography would also qualify by these
criteria.
12 Chen & Gardner, 2005.
Gardner uses these criteria somewhat
loosely, for most but not all of his intelli¬
gences satisfy all the criteria. Here are the
intelligences that he has identified:
Linguistic intelligence: This refers to skill
with verbal and written language.
Writers and poets are prototypes of
individuals possessing this skill.
Logico -mathematical intelligence: Skill in
mathematical, numerical, and abstract
logical reasoning. Mathematicians and
computer programmers are offered as
prototypes.
Spatial intelligence: A skill possessed by
graphic designers and architects. It is
the ability to perceive and manipulate
spatial and visual images.
Bodily /kinesthetic intelligence: The skills
displayed by dancers and athletes.
They deal with the ability to control
bodily movements.
Naturalistic intelligence: This skill is criti¬
cal for archeologists and botanists, two
rather different fields. It involves skill
in dealing with elements in the natural
environment.
Interpersonal intelligence: The skills poli¬
ticians have to have! The skills re¬
quired to deal with people, as opposed
to ideas or things.
Intrapersonal intelligence: Skill in under¬
standing and regulating one’s own
emotions, strengths, and desires.
These are certainly desirable traits, but is
it useful to call all of them “intelligence”?15
Gardner himself has admitted that his use
of the term “intelligence” was a calculated
attempt to capitalize on what he saw as a
tendency in Western culture to value men¬
tal competencies, rather than an attempt to
further the study of what others had defined
as “intelligence.”14 Does grouping these var¬
ious traits together lead to a natural kind, a
set of behaviors that have similar causes and
lead to similar effects?
13 Hunt, 2004.
14 Gardner, 1999, p. 33.
TAKING INTELLIGENCE BEYOND PSYCHOMETRICS
n7
5.1.2. The Evidence for the Theory
of Multiple Intelligences
MI theory resonates with popular ideas
about intelligence. The idea that there are
multiple talents and that people can be high
on one (e.g., artistic skill) while low on oth¬
ers (e.g., analytic skills) accords with pop¬
ular conceptions of the distribution of tal¬
ent. Surveys have consistently shown that
the popular concept of intelligence, in both
developed and undeveloped countries, is
broader than the range of talents evaluated
in tests. Social skills, in particular, are given
more weight in the popular mind than in
the classic testing paradigm. When people
are asked to rate themselves and others on
Gardner’s multiple intelligences, they see
the task as reasonable. But when people are
asked to give ratings of overall intelligence
this is also seen as a reasonable task, and ver¬
bal, mathematical, and spatial skills are the
biggest contributors to the overall ratings.15
Whether or not you get popular acceptance
of the MI model or of something closer to
the g-VPR model depends a good deal on
how you phrase the question.
In any case, popular opinion is hardly sci¬
entific evidence. We need a source of facts,
not opinion. Evidence for the utility of a the¬
ory of intelligence comes from three sources:
the theory’s utility in explaining displays of
human abilities both in the school and in the
workplace, and from psychometric studies
intended to evaluate the theory.
The MI model has been an easy sell to
educators. They have developed a num¬
ber of educational programs based on the
multiple intelligences, usually for the early
childhood and primary school years. The
prototypical program first assesses children,
and then provides educational interven¬
tions designed to develop their strengths.
The intervention is followed by a final
assessment. Consistent with Gardner’s view,
assessment is typically done by engag¬
ing children in a (hopefully) interesting
task, rather than by formal testing. These
15 Fumham, 2001.
programs have been reviewed favorably by
Gardner himself,16 and with somewhat more
reservation by some of the authors in an
edited volume examining evidence for the
theory.17
These studies show that children in kin¬
dergarten and the early elementary school
years become more engaged when they par¬
ticipate in activities that suit their individ¬
ual talents, than when they must conform
to activities chosen for a class as a whole.
Not surprisingly, these talents are then fur¬
ther developed, for you learn to do what you
practice doing.
But how relevant is this to the develop¬
ment of intelligence? Or, for that matter, to
the design of school programs? To answer
these questions we would have to compare
the later performance of children who had
gone through one of the MI programs to the
later performance of comparable children
who had gone through a standard early edu¬
cation program. To my knowledge no such
test has been reported. I also worry that if
such an experiment did not provide support
for the popular Multiple Intelligence model
the report would suffer from the bias against
reporting negative results.
I doubt that negative results would have
much influence, for Gardner and his sup¬
porters can offer two plausible explana¬
tions for them. Gardner claims that higher
levels of instruction (e.g., middle school,
high school, and college) are organized
toward the development of traditional ver¬
bal and analytic skills. Therefore, the results
obtained by training other intelligences dur¬
ing the primary school years would not
be valued as children progressed up the
educational pipeline.18 A second reason for
16 Gardner, 1993a; Cheng & Gardner, 2005.
17 Schaler, 2006.
18 A good case can be made for the argument that
schools are oriented toward developing analytic and
verbal skills, at the expense of developing skills in
artistic and social endeavors. But is this a bad thing?
Are schools supposed to develop children's varied
talents, or are they supposed to train a workforce for
a post-industrial society? Educators tend to take the
first view. Many observers from outside the edu¬
cational field take the second view. The National
n8
HUMAN INTELLIGENCE
disregarding negative results is that a pro¬
gram designed to develop multiple intelli¬
gences may not have been properly orga¬
nized. Gardner has said that he approves of
some implementations, and not of others.
This offers him an out; if a Multiple Intel¬
ligences program does not show success, it
must be because the program was improp¬
erly implemented. And of course this could
be true. But no definition of proper imple¬
mentation has been given other than that the
results should be in accord with the theory.
As far as I know, there has never been a
rigorous attempt to evaluate the Multiple
Intelligence approach as a guide to selec¬
tion or evaluation of personnel in workplace
settings. Indeed, the evidence is strongly in
favor of the use of general measures of intel¬
ligence - that is, ^-loaded measures - as
predictors of workplace performance. This
is not to say that special skills are irrel¬
evant in certain extreme situations. Poets
and airplane mechanics do different things!
However, the advocates of MI have been
primarily concerned with education.
Three Canadian psychologists, Beth
Visser, Michael Ashton, and Philip A, Ver¬
non, took a psychometric approach to the
evaluation of MI theory.19 They constructed
two different behavioral tests for each of the
eight intelligences. For example, linguistic
ability was tested by a vocabulary test and
by a test requiring people to identify a word
that had the opposite meaning from a tar¬
get word. Social intelligence was assessed
by a task in which participants saw sev¬
eral cartoon drawings depicting part of a
story and then had to choose a final cartoon
indicating a logical completion of the story,
and by a task in which participants had to
determine what a word or phrase meant,
in context. In addition to tests that evalu¬
ated the intelligences specified by MI, par¬
ticipants took the Wonderlic Personnel Test
Academy of Science (NAS) report Rising Above
the Gathering Storm: An Agenda for American Sci¬
ence and Technology (2007) can be read in places as
expressing concern that our system is training too
many poets, authors, and business managers, and
not enough engineers and scientists.
19 Visser, Ashton, & Vernon, 2006.
(WPT) test, which is a g-loaded test that
stresses rapid responding to simple prob¬
lems. (See the description of this test in
Chapter 2.)
The 17 tests were given to 200 adult vol¬
unteers drawn from a university commu¬
nity. Participants were recruited from stu¬
dents in thirty different departments, thus
maximizing the chance that the sample
would include people with diverse sets of
talents.
The Visser and colleagues study stacked
the deck in favor of the MI model. The par¬
ticipants were adults with different train¬
ing and interests, and were of above-average
general intelligence. As noted in Chapter 4,
greater differentiation of ability is typically
found in above-average samples than in
below-average samples. Nevertheless, the
results did not support the model. The tests
of what everyone would agree are cognitive
abilities - the linguistic, mathematics, and
spatial tests - had strong loadings on a gen¬
eral factor and substantial correlations with
the WPT. So did the tests of natural intel¬
ligence, which evaluated a person’s knowl¬
edge about the biological world. This result,
and other results in the literature,20 are con¬
trary to Gardner’s frequent denial of the
strength of the general factor in intelligence.
The tests of noncognitive factors, such as the
tests of bodily/kinesthetic intelligence, had
low loadings on the general factor. This is
consistent with the argument that combin¬
ing all these variables under the name “intel¬
ligence” forms a class of behaviors that is no
longer a natural kind.
A second result was also damaging to
the MI theory. According to MI theory the
correlations between different tests evalu¬
ating the same intelligence should be high.
For instance, after accounting for the effects
of general intelligence there should be a
high correlation between the two linguis¬
tic tests, the two mathematical tests, and so
on. In general, there was not. The two lin¬
guistic tests had a residual correlation, after
20 Brody, 2006. See also Carroll’s (1993) discussion of
lack of evidence for Gardner’s view of general intel¬
ligence.
TAKING INTELLIGENCE BEYOND PSYCHOMETRICS
119
removing the effects of g, of .29. None of
the other seven correlations between tests
of the same intelligence exceeded .18, and
the average correlation was .10.
In sum, this study provided no evi¬
dence whatsoever that MI theory actually
describes the distribution of cognitive skills
in college students.
Gardner21 dismissed these results, on two
grounds. He disputed the appropriateness of
some of the tests, and he rejected the idea
that it is even appropriate to evaluate the
theory using the standard psychometric test¬
ing paradigm. He concluded that Visser and
her colleagues got the results that they did
because they recreated the standard scholas¬
tic testing situation, with a strong emphasis
on logical reasoning.
Gardner’s rejoinder to Visser and her
colleagues was typical of his reaction to
attempts to assess his ideas. In his own
words,
As I have often explained except for Project
Spectrum I have not devoted energies to the
devising of tasks that purport to assess ML I
have no objection to others doing so though
the efforts so far have been modest. In my
own experience, I have been impressed by
efforts to create environments in which the
use of multiple intelligences is highlighted.
Gardner, 2006b, p. 504
5.1. 3. An Evaluation of MI
Despite the popularity of the MI idea in
some educational circles, there is virtually
no objective evidence for the theory. The
attitude toward measurement that Gardner
expressed in his reaction to Visser and her
colleagues does not lead to good science. Sci¬
entific ideas are tested by their ability to
account for data, and such tests cannot be
carried out unless there is a commitment
to means of measurement. Gardner cannot
simply dismiss results such as those of Visser
and her colleagues unless he offers an alter¬
native means of measurement.
The fact that children can be enthusi¬
astic about a program that plays to their
21 Gardner, 2006b.
own task-assessed strengths is interesting,
although fairly obvious. However, there
appears to be little evidence, positive or
negative, that early childhood training using
Multiple Intelligences theory translates into
good performance later on in the educa¬
tional process. One can blame the educa¬
tional process for this, arguing that the later
educational stages stress formal, analytic rea¬
soning to the detriment of other intelli¬
gences. Whether such an attack is valid or
not depends upon what one thinks formal
education is for. (See footnote 18 for an
expansion on this idea.)
There is no evidence that expanding
the notion of intelligence beyond obvious
cognitive characteristics, to include such
things as musical and bodily/kinesthetic
“intelligence,” creates a natural kind.
These noncognitive characteristics are all
admirable talents. They can and should
be studied separately. Forcing them under
the umbrella of intelligence will lead to an
unmanageable topic of study. It has been
said that science proceeds when it defines
tasks that carve nature at its joints. If this
is true, Multiple Intelligence theory is not a
map for scientific progress.
Having said this, I close with a strong
endorsement of one of Gardner's positions,
as being useful for science, and a more qual¬
ified endorsement of another, as a way of
thinking about education.
Gardner makes a good point when he
advocates expanding the evaluation of intel¬
ligence beyond the conventional psycho¬
metric testing paradigm. The field could
definitely profit from more analyses of indi¬
vidual differences in performance of every¬
day tasks, in fields ranging from the evalua¬
tion of surgeons, mechanics, and lawyers, to
the observation of schoolchildren. Indeed, in
industrial-organizational psychology there
is considerable concern that personnel
assessment is limited by our ability to
record details of on-the-job performance.22
Gardner deserves credit for highlighting
the problem, although (possibly due to his
22 S. Hunt, 2007.
120
HUMAN INTELLIGENCE
disdain for assessment) he has not provided a
solution.
The qualified endorsement is that any
humane person ought to agree with Gard¬
ner that schools should, insofar as possible,
encourage students to develop their personal
abilities and interests. I would like to see
students given much more opportunity to
develop talents in music, the performing
arts, and even, for all students rather than
just for selected stars, athletics. However, I
also want students to be prepared for the
advanced, post-industrial society that they
are going to live in. In order to meet this
goal schools have to stress the development
of linguistic and mathematical skills. The
extent to which other topics can be taught
depends on how much time and money soci¬
ety is willing to invest in education. Defining
the curriculum is a matter for educational
policy, not for science.
5.2. Robert J. Sternberg's Theory
of Successful Intelligence
The next theorist whose work will be dis¬
cussed, Robert J. Sternberg, has, like Gard¬
ner, attempted to expand our views of
intelligence beyond psychometrics. Unlike
Gardner, though, Sternberg’s research is
grounded in previous research, and he has
assiduously conducted experiments to sup¬
port and expand his ideas.
Sternberg has been a strong advocate for
the adoption of his theories in education
and in industry. To that end he has writ¬
ten widely and made many presentations
to nonspecialists, including both the gen¬
eral audience and policy makers. By con¬
trast, most developers of psychometric theo¬
ries have written mainly for other scientists.
When evaluating Sternberg's work one has
to consider both what he is saying, either
about his own work or that of others, and
the audience to whom he is saying it, either
specialists or nonspecialists.
I will first consider Sternberg's views on
the work of others. This provides a motiva¬
tion for the development of his own theo¬
retical position.
5.2.1. Sternberg's Criticism of Previous
Work on Intelligence
Sternberg has shown a great deal of aware¬
ness of previous work on intelligence. This
is evidenced by his excellent review of the
field as it stood in 199c.23 In it he pre¬
sented a carefully balanced analysis of differ¬
ent theoretical approaches to intelligence, as
they existed at that time. He was dissatis¬
fied with the field because of what he saw
as its myopic view of the range of human
intelligence. He also felt that an appropri¬
ate goal for intelligence research was to pro¬
vide useful information to guide educators
and industrial leaders in the development of
people's capacities, rather than simply cata¬
loging individual differences. To satisfy this
goal he has written numerous books and
articles directed to the general public and
to educators. Some of these contain harsh
criticisms of the field, i have already quoted
from one of these books. Here is what he has
to say on the next page of the same work.
Almost everything you know about intelli¬
gence - the kind of intelligence psychologists
have most often written about - deals with
a tiny and not very important part of a
much broader and more complex intellec¬
tual spectrum.
Harsh words indeed] In his writings in the
technical literature Sternberg takes a more
measured tone. For instance, he has been
very much involved in developing educa¬
tional programs that combine testing with
educational programs that use the tests to
tailor instruction to student strengths and
weaknesses.24 Here is what he has to say
about the results of a large project involv¬
ing an attempt to improve the prediction
of success in college (after citing several
meta-analytic studies of prediction of col¬
lege performance, which produced valid¬
ity estimates for predicting freshman GPA
ranging from .4 to .6):
All together these results suggest good pre¬
dictive validity for the SAT for freshman
23 Sternberg, 1990.
24 Sternberg, Grigorenko, & Zhang, 2008.
TAKING INTELLIGENCE BEYOND PSYCHOMETRICS
121
college performance. But as is always the
case for any test or type of test, there is
room for improvement. . . . Thus the theory
[of successful intelligence] does not suggest
replacing, but rather augmenting the SAT
in the college admissions process.
Sternberg, 2006, p. 322; emphasis in
the original text
This is a correct analysis of the cur¬
rent situation, and a reasonable proposal for
expanding the study of intelligence. What is
the nature of the expansion Sternberg pro¬
poses?
5.2.2. The Theory of Successful Intelligence
Because Sternberg has written so much,
and because his thinking has (appropriately)
evolved over his career, it is sometimes dif¬
ficult to assess just what his current posi¬
tion is. My comments will be based largely
on the arguments presented in his book
Practical Intelligence in Everyday Life,zt> pub¬
lished in 2ooo; and subsequent related pub¬
lications. Practical Intelligence is a narrative
account of research conducted over the pre¬
vious twenty years, and thus provides a
good jumping-off place for an analysis of his
ideas.
Sternberg distinguishes three classes of
intelligence: analytic (sometimes called aca¬
demic), creative, and practical intelligence.
He believes that conventional psychomet¬
ric and personnel screening tests, such as
the SAT and AFQT, are essentially mea¬
sures of analytic intelligence. The analytic
intelligence tests that he and his colleagues
have developed intentionally resemble con¬
ventional cognitive aptitude tests. They do
not seem to have the predictive power of
tests like the SAT and AFQT, but that
could be because they are much briefer and
because, after all, the conventional tests now
in use have been refined over the course of
decades. Sternberg claimed that any major
expansion of our ideas of intelligence will
require the development of tests of creative
and practical intelligence.
25 Sternberg et al., 2000.
CREATIVITY
There is a long history of trying to develop
tests of creativity apart from tests of intelli¬
gence. This has proven to be a difficult task,
for a simple reason. What is the criterion?
It is easy to collect anecdotes about unde¬
niably creative individuals, especially those
with very high levels of accomplishment.
Albert Einstein is a frequently cited exam¬
ple. However, it is difficult to go further
because highly creative people are rare.
Dean Simonton, a professor at Univer¬
sity of California, Davis, who has made an
extensive study of high levels of creativity,
has pointed out that, unlike most human
traits, creativity is not normally distributed.
The majority of truly creative contributions
are made by a very small fraction of the
workers in any given field. (If a contribu¬
tion is commonplace it will probably not be
considered creative.) Simonton also points
out that creative contributions are usually
specialized. This could be because the psy¬
chological traits required to be creative are
specific to a particular field. There is no
compelling argument that being a creative
chemist and being a creative artist draw on
the same traits. It could also be that only a
tiny, tiny percentage of people are creative
in more than one field because there sim¬
ply is not enough time. There is very little,
if any, evidence for the existence of effort¬
lessly creative people, who can contribute to
one field after another without working very
hard in any of them. And finally, both cre¬
ative accomplishments and their recognition
depend very much upon the social setting.26
These considerations make it difficult to
draw generalizations from the study of peo¬
ple who we all agree are creative. Therefore,
scientists who want to study creativity often
wind up studying people who are believed
to be creative in everyday life. This is not
easy, either. Extrapolating from Simonton’s
analysis, the difficulty may be that there are
not that many such people. Or it might be
that they are there, but are not motivated to
reveal themselves, because we do not want
to have to be creative on a daily basis. The
26 Simonton, 1984.
122
HUMAN INTELLIGENCE
world works much better if, most of the
time, the problems that people encounter
are ones that they know how to solve. A
day on which none of the problem-solving
methods you have learned works is a bad,
frustrating day.
This presents a problem for a test con¬
structor. Psychologists construct intelligence
tests by identifying people who display dif¬
ferent levels of cognitive competence, as
indicated by grades, supervisor ratings, and
objective tests of on-the-job performance,
and then designing tests that discriminate
between performers and nonperformers. If
creative behaviors themselves are rare or
hard to identify, test constructors do not
know where to begin.
It is possible to measure a sort of creativity
referred to as divergent thinking. Examinees
are shown a situation or object, and asked
to list as many ways the test stimulus can be
used or interpreted as they can - for exam¬
ple, asking people to list the different ways
you can use a brick. The response is then
assessed by judges. Such tests are face-valid
as measures of the ability to produce many
and/or unusual responses. But does this have
anything to do with the creation of creative
products? More particularly, does a creativ¬
ity test requiring divergent thinking measure
a useful trait that is not measured by intelli¬
gence tests?
There is evidence that it does. E. Paul
Torrance, a professor of psychology at the
University of Georgia, developed a widely
used series of creativity tests along the lines
just described. During the 1960s they were
given to elementary school children in two
selected schools. The students were also
given the Stanford-Binet intelligence test.
Torrance’s participants were quite bright;
the mean IQ was 120. Creativity was assessed
by records of performance in adulthood, in
some cases as much as forty years later.
A modern analysis of Torrance's data27 has
shown that in this high-IQ sample creativity
measures taken in childhood made a sub¬
stantial contribution to predicting lifetime
27 Plucker, 1999.
creativity, independently of predictions
based on childhood IQ scores.
Sternberg is on firm ground in calling for
an expansion of intelligence testing to con¬
sider creativity. Would the effort be worth
the expense? That is hard to say. Statisti¬
cally, the gains in prediction would probably
be small. However, identifying creative peo¬
ple could be extremely important for eco¬
nomic and social reasons.
PRACTICAL INTELLIGENCE
Practical intelligence refers to the ability to
deal with ongoing, realistic, and at times
ill-defined problems. Sternberg has placed
great stress on the importance of measuring
this trait. He claims that tests of academic/
analytical intelligence are incomplete
because they rely on abstract problems
in which all necessary information is pre¬
sented. The practical intelligence problems
that Sternberg and his colleagues have
developed either ask for specific, task¬
relevant knowledge or describe situations
claimed to be realistic, and ask examinees
what they would do.
Table 5.1 shows some examples of ques¬
tions that have been asked on tests of
practical intelligence. Many of these ques¬
tions resemble current industrial personnel
selection practices. Asking for an analysis
of a hypothetical situation is a low-fidelity
simulation of performance on the job. In
industrial-organizational psychology, tests
that do this are called situational judgment
tests, and have been used for years. It is of
interest that within industrial-organizational
psychology they are considered to reveal
personality traits, such as preferences for
certain types of action, as well as contain¬
ing a cognitive component.28
Sternberg and his collaborators have
stressed the importance of tacit knowledge
as a component of practical intelligence. By
this they mean knowledge that is not explic¬
itly taught but that is required in many sit¬
uations. Tacit knowledge is often procedu¬
ral knowledge, knowing what to do, rather
28 S. Hunt, 2007, pp. 69-70.
TAKING INTELLIGENCE BEYOND PSYCHOMETRICS
123
Table 5.1. Examples of tasks used by Sternberg and his colleagues to assess practical
intelligence
Setting Item Type Item Reference
Evaluating
knowledge of folk
medicine in rural
Kenya.
Prediction of
success of graduate
students in an
American business
school.
Prediction of
graduate student
success in an
American business
school.
Knowledge of the
uses of traditional
remedies.
Skills in solving a
practical problem.
Situational
judgment test.
The examinees (Kenyan Sternberg et al.,
children] are given a 2001.
description of a person's
symptoms, the Kenyan name
of the illness, and asked which
of five native herbal medicines
are appropriate for the case.
The examinee is given the role Hedlund et al.,
of the human resources 2005.
manager of a manufacturing
plant facing a personnel
shortage. Employees are
working excessive amounts of
overtime and morale is low.
The examinee is to suggest a
course of action. Materials
provided: current employment
figures and job-satisfaction
survey results.
The examinee is presented Hedlund et al.,
with a personnel problem 2005.
similar to the one just
described, but with
considerably more detail
given. The examinee is then
asked to rate several possible
solutions, including hiring
temporary workers, hiring
full-time employees but
warning them that they may
be let go if product demand
drops, mixing the two
solutions, or letting each
division within the plant
decide upon its own solution.
than declarative knowledge, knowing how
to describe the situation. This is a notion
worth examining.
The environment offers different degrees
of support for acquiring knowledge, rang¬
ing from presenting examples offered with¬
out comment to presenting ideas in formal
instruction, either in or out of school. Peo¬
ple will vary in the extent to which they
can articulate how or why they take cer¬
tain actions. For instance, most people learn
sports like bicycling and skiing by being
coached. However, relatively few skilled
skiers and bicyclers become good coaches,
in part because they cannot verbalize what
the novice is supposed to do. The concepts
of tacit knowledge and explicitly instructed
knowledge represent end points on a contin¬
uum, rather than mutually exclusive classes.
In the narrowest sense, tacit knowledge
should be knowledge that people acquire
by observation and experience, without any
124
HUMAN INTELLIGENCE
form of instruction. As an amusing exam¬
ple, think of the way that pre-adolescents
learn to curse. The learning phase usually
includes some hilarious misuses of improper
language.
At times Sternberg and his colleagues
write as if tacit knowledge were identical to
practical knowledge. This blurs an impor¬
tant distinction. Much practical knowledge
is acquired through formal instruction. For
example, airline captains have a great deal of
job-relevant knowledge about how to fly air¬
craft. Much of it is acquired in flight school.
As late as the 1960s Polynesian navigators in
the Marianas Islands went to formal schools
to learn how to sail outrigger canoes across
vast reaches of the Pacific.29 Simply showing
that knowledge is practical does not ensure
that it has been acquired tacitly.
This restriction does not deny the impor¬
tance of tacit knowledge. Most apprentice¬
ship learning is tacit. The master and the stu¬
dent have a problem to solve, and the master
solves it, with the assistance of the student,
who, hopefully, will later be able to act on
his or her own. During problem solving the
master may or may not explicitly instruct
the student, but what instruction there is
is almost always in the context of practical
problem solving.
The research issue is whether differ¬
ent cognitive skills are required for learn¬
ing by observation, apprenticeship learning,
and formal classroom learning. It could also
be that general intelligence is a substantial
requirement for all forms of learning. The
question is one to be settled by investiga¬
tion, not by arguments that one or the other
conclusion must be true.
5.2.3. The Evidence for the Theory
of Successful Intelligence
Validation of Sternberg’s theory requires
that tests be defined for each of the three
abilities, that the abilities be substantially
uncorrelated (for otherwise they might
simply be reflections of the pervasiveness
of general intelligence), and that the
29 Hutchins, 1983.
uncorrelated portions of the tests of aca¬
demic, creative, and practical intelligence
predict important behaviors both within and
outside of an academic environment. In the
book Practical Intelligence in Everyday Life,
and in several publications since then, Stern¬
berg has claimed that these criteria have
been met.30 This claim has been disputed.31
It would not be possible to go through point
and counterpoint of every study. I have
selected a few examples to highlight the
issues involved.
Three studies are relevant to the predic¬
tion of academic knowledge in three differ¬
ent educational settings.
In one study Sternberg and his colleagues
administered tests of analytic, creative, and
practical intelligence to over three hundred
students attending a pre-college preparatory
course at Yale University. At the time the
students were also taking an introductory
psychology course. This made it possible
to construct a three-by-three experimental
design, in which students were categorized
as having strengths in analytic, creative,
or practical intelligence, and then were
assigned to study sections where the instruc¬
tion stressed analytic, creative, or practical
problem solving. Sternberg claimed that stu¬
dents performed best when the method of
instruction matched their strengths in the
appropriate type of intelligence.32
Nathaniel Brody, the author of several
textbooks and a number of scholarly com¬
mentaries on intelligence, made several crit¬
icisms of this study.33 He noted that the
sample had been pre-selected to be high
on general intelligence, for they were, after
all, students intending to enter an extremely
selective university. The factorial experi¬
ment supporting the crucial aptitude x treat¬
ment interaction dealt with only 199 of the
324 students in the study. When the entire
30 Sternberg et al., 2000. See also his reply to criti¬
cisms (Sternberg, 2003b) and a claim for the utility
of his measures in the selection of college students
(Sternberg, 2006, 2007) and in general education
(Sternberg, Grigorenko, & Zhang, 2008).
31 Gottfredson, 2003a; Hunt, 2008.
32 Sternberg et al., 1996, 1999; Sternberg, Grigorenko,
& Zhang, 2008.
33 Brody, 2003.
TAKING INTELLIGENCE BEYOND PSYCHOMETRICS
125
sample was considered there was no evi¬
dence of an aptitude by treatment interac¬
tion. Finally, Brody reanalyzed Sternberg’s
data and concluded that the predictive abil¬
ity of the three allegedly independent tests
of different types of intelligence was due
solely to a common trait that underlay them
all. Brody felt that this represented general
intelligence.
Sternberg replied to Brody by saying that
he regarded the Yale study as a preliminary
one, that he and his colleagues had devel¬
oped better tests of creative and practical
intelligence, and that positive results would
shortly be forthcoming.34 Sternberg subse¬
quently published the results of a very large-
scale study, the RAINBOW study, involv¬
ing a number of universities and colleges,
in which an attempt was made to predict
first-year grade point averages (GPAJ from
a combination of SAT scores and scores
on tests of analytic, creative, and practical
intelligence.35 The RAINBOW study cov¬
ered a diverse set of institutions, ranging
from California community colleges to Ivy
League universities. One of the tests of cre¬
ativity did improve predictivity beyond the
predictivity achieved using the SAT alone,
but the tests of practical ability did not add
anything beyond predictions using the SAT
alone.
In a nontechnical note addressed to col¬
lege administrators, Sternberg claimed that
predictivity was increased by 15%, in terms
of variance accounted for. This is a sub¬
stantial gain indeed.36 Since the SAT pre¬
dicts 10-15% of the variance in performance
of accepted students, if we accept Stern¬
berg’s statement at face value, a test pro¬
gram that incorporated Sternberg’s creativ¬
ity tests along with the SAT or ACT could
predict somewhere between 15% and 25% of
the variance in first-year GPA.
Given this claim, a close look at the data
supporting it is in order.
The added predictivity was due to a sin¬
gle test of creative ability, in which exami-
34 Sternberg, 2003b.
35 Sternberg, 2006.
36 Sternberg, 2007.
nees saw a collection of pictures, then made
up and told the examiner a story about the
pictures. Thus the test was rather like view¬
ing comic book pictures without the cap¬
tions, and then telling the story. Surpris¬
ingly, there was no gain in predictivity if the
examinees had to write the story. Asserting
that "Creative Intelligence” added to predic¬
tivity is something of an overgeneralization,
since one test worked and the other did not.
More generally, it is difficult to interpret the
RAINBOW results because of a statistical
problem.
An attempt was made to establish a rela¬
tionship using multiple tests of both prac¬
tical and creative intelligence, and one of
them worked. This is a reasonable model for
exploratory research, but it opens the gates
to capitalization on chance results. In such
cases the only thing to do is to attempt to
replicate the findings.
Because a result could have been obtained
by chance does not mean that it was. The
result is consistent with previous research on
creativity, especially Torrance's work, cited
earlier. According to Sternberg the RAIN¬
BOW project is to be expanded.37 If the
results do replicate, a good case will have
been made for including tests of the ability
to produce fictional material as an evalua¬
tor of college readiness. 38 This is an impor¬
tant theoretical finding. Whether or not this
means that Sternberg has provided a theo¬
retical case for Creative Intelligence, apart
from the sorts of testing of linguistic com¬
petence that occur in conventional testing,
is a somewhat different question.
In order to make a case for creativity
as a predictor Sternberg and his colleagues
will have to show that several creativity
measures predict success in college, beyond
the prediction that can be obtained using
37 ibid.
38 The most recent versions of the SAT include a test
of writing ability. If Sternberg and colleagues’ results
depend upon verbal fluency, they should not add to
predictivity beyond that obtained by the expanded
SAT. If the key point is creativity, however, then
their tests should be unique predictors, because the
SAT writing requirement is, in terms of content,
rather prosaic. Grading is for English usage rather
than for literary merit.
126
HUMAN INTELLIGENCE
standard tests, such as the SAT and ACT.
Ideally there should be some nonverbal cre¬
ativity measures, to avoid confounding cre¬
ativity and verbal performance.
The RAINBOW study failed to sup¬
port Sternberg’s theorizing in one important
way. None of the measures of practical intel¬
ligence in an academic setting added to the
amount of prediction that could be obtained
with the SAT alone. This finding is consis¬
tent with Gottfredson’s claim39 that tests of
practical intelligence are actually evaluating
general intelligence (g) rather than evalu¬
ating a separate dimension of ability, as is
claimed in Sternberg’s Theory of Successful
Intelligence.
Recognizing this and similar concerns,
Sternberg and his colleagues have acknowl¬
edged that the Theory of Successful Intel¬
ligence provides one among several possible
ways of motivating the empirical studies and
organizing their results with respect to aca¬
demic accomplishment.40
The RAINBOW study attempted to do
many things, using a complicated design.
It is useful to look at a more focused
study, in which Sternberg and his colleagues
attempted to predict the performance of
students in their first year of a Master of
Business Administration (MBA) program at
the University of Michigan.41
Students in the Michigan program took
two tests of practical intelligence - a test
of their ability to analyze and recommend
actions on realistic business problems, and
a situational judgment test similar to the
one described in Table 5.1. An attempt
was made to predict their first-year course
grades, either using the two tests and the
students’ scores on the Graduate Manage¬
ment Admissions Test (GMAT), which they
had taken prior to entry, or using only
the GMAT. The tests improved prediction
over the predictions based on the GMAT
alone. This finding is clearly what would
be predicted by the Theory of Successful
Intelligence.
39 Gottfredson, 2003a, b.
40 Sternberg, 2006, p. 322.
41 Hedlund et al., 2006.
Sternberg and his colleagues have com¬
pleted a number of studies outside of aca¬
demic settings. These include both stud¬
ies of industrial performance and studies
of intelligence in developing countries. A
study of military leadership is of particular
interest, both for its own importance and
because it is illustrative of potential indus¬
trial applications.42
The first part of the study consisted of an
extensive review of military leadership skills,
including examinations of leadership duties
at three different military levels: platoon
leaders, company commanders, and battal¬
ion commanders. The distinction is impor¬
tant because the type of leadership required
differs across these levels.
Platoon leaders lead by direct, face-to-
face contact with from twenty-five to forty
soldiers. Platoon leaders are almost invari¬
ably lieutenants, recently commissioned
officers. They are assisted by three to five
sergeants, noncommissioned officers who
have had less formal command training than
the lieutenants, but may have had consider¬
ably more experience in the military.
Company commanders are usually cap¬
tains or senior first lieutenants. In the Amer¬
ican services they will have had roughly
five to ten years’ experience in the military.
Depending on the type of company, they
will command from one to two hundred sol¬
diers, organized into three or four platoons.
The company commander has a small staff,
including an executive officer (usually a lieu¬
tenant) and a few senior sergeants. His job
mixes administrative duties with a good deal
of face-to-face leadership of platoon leaders
and sergeants.
Battalion commanders are majors or,
more likely, lieutenant colonels. They will
have had at least ten years’ experience as
officers. Depending upon the type of bat¬
talion, they will be in command of any¬
where from three hundred to six hun¬
dred soldiers, organized into companies or
analogous units. Battalion commanders are
essentially administrative leaders. They are
assisted by staffs containing experienced
42 Hedlund et al., 2003.
TAKING INTELLIGENCE BEYOND PSYCHOMETRICS
127
officers and senior sergeants, who develop
plans and relay orders to company com¬
manders and, rarely, individual platoon
leaders.
Sternberg and his colleagues developed
three measures of what they describe as tacit
knowledge of military leadership for each
level of command. These covered a num¬
ber of practical aspects of leadership, rang¬
ing from motivating subordinates to com¬
municating with senior officers. In addition
they utilized a standard measure of verbal
reasoning (similarities and analogies), and a
measure of tacit knowledge developed for
use with civilian managers. The officers were
rated for their leadership ability by their
superiors, peers (except for battalion com¬
manders), and subordinates (except for pla¬
toon leaders).
The question of interest is whether the
measures of military tacit knowledge pre¬
dicted leadership ratings, over and above the
prediction possible from use of the verbal
reasoning measure. A secondary question is
whether the measures designed for the mil¬
itary are more accurate predictors of lead¬
ership ratings than the measure derived for
use in civilian settings. According to the the¬
ory, the military measures should be more
accurate because they evaluate knowledge
required by the local situation.
The results presented a mixed bag. The
military knowledge tests were the best pre¬
dictors of leadership ratings, but no test
did particularly well. The six possible cor¬
relations between the military knowledge
questionnaires and rated leadership ranged
from .46 to —.11. What is even more puz¬
zling is the pattern of correlations. The .46
represented a positive relation between the
military knowledge test and the effective¬
ness ratings of battalion commanders by
their superior officers. The —.11 represents
a relationship between the same test and
the ratings the same battalion commanders
received from their subordinates! Similar
but not quite so puzzling anomalies were
found in the ratings of platoon leaders and
company commanders. A further statistical
analysis showed that knowledge of military
leadership did have some predictive validity
beyond the other measures. However, the
effects were far from spectacular. Only three
of the seven possible comparisons were sta¬
tistically reliable, and only one, prediction
of superiors’ ratings of battalion comman¬
ders, was large enough to be of practical
importance.
As in the case of the studies of academic
performance, this is the sort of study that is
suggestive, but replication is required before
strong conclusions can be made.
The studies in non-Western cultures
include studies of a pastoral/farming com¬
munity in rural Kenya, health practices in a
Russian city, and hunters in an Alaskan Inuit
community.43
The Alaskan study44 provides typical
results. The participants were Yu’pik Inuit
adolescents residing in several small towns.
The people in this area are best described as
semitraditional; they live in settled towns,
go to Alaskan state schools, and participate
in the normal American economy. They
also do a considerable amount of hunting
and, to some degree, follow traditional social
customs. Scores on a test of knowledge of
hunting lore and Inuit terms were largely
independent of scores on conventional tests
of intelligence. The scores on the tests of
Inuit knowledge correlated with estimates
of hunting skills. Similar results were found
in a study in Kenya, where the targeted
information was knowledge of traditional
medicine.
Both these studies illustrate a simple prin¬
ciple. People will learn to solve the cognitive
challenges their societies present. There will
be individual differences in the amount of
learning, and such differences will predict
performance in that society.
5.2.4. Evaluation and Critique
of the Theory
Sternberg’s approach can be evaluated on
three grounds; his criticisms of other the¬
ories, the logic of his own theory, and the
evidence he offers for it.
43 Sternberg, 2004.
44 Grigorenko et al., 2004.
128
HUMAN INTELLIGENCE
CRITIQUE OF THE ATTACK
Sternberg’s attacks on other theoreti¬
cal approaches are excessive, especially
when he writes for nontechnical audiences.
“Everything people have been told about
intelligence” is not false, and intelligence
tests do not solely evaluate “narrow abilities
of use only in school settings.” Fie makes
a case against conventional tests by offer¬
ing anecdotes about failures in prediction,
including some of his personal experiences.
This is an effective rhetorical device, but it
is not a valid argument, unless he is argu¬
ing against a claim that the tests are per¬
fect predictors of performance. No one has
ever made such a claim. The claim that is
made, and supported by massive evidence, is
that the tests, although far from perfect pre¬
dictors of performance, are the best predic¬
tors that we have, both in academic and
industrial settings.45
Nonetheless, Sternberg has a serious
point. Since Binet’s time psychometricians
have seen their task as one of measur¬
ing intelligence, and then simply reporting
the result. There has been a great deal of
research on increasing test validity and mak¬
ing test administration more efficient. What
has been missing is concern for what hap¬
pens after test scores have been obtained.
There is an implicit (and often not-so-
implicit) assumption that personnel decision
makers will establish a cut score, and then
either accept or reject candidates. Reality is
more complex. Educators have to do some¬
thing with the students they have; industrial
supervisors have to design jobs and train¬
ing programs suitable for the available work¬
force; and military commanders have to lead
the troops who have enlisted. What Stern¬
berg has done is to move away from a pure
selection model for the use of test scores to
a model where test scores are meshed with
training and educational decisions.
This is not a novel idea, for industrial-
organizational psychologists have been
doing this for some time, especially in
45 For reviews, see Gottfredson (1997], Hunt (1995),
and Schmidt & Hunter (1998). The topic is explored
in Chapter 10.
military settings. For instance, in addition
to the AFQT, the ASVAB battery provides
scales indicating skills or knowledge in spe¬
cial fields (e.g., electronics). These scales
are then used, along with other criteria, to
assign soldiers to training programs for var¬
ious military specialties. In educational and
most civilian industrial applications, though,
it is much more common for test scores to
be used to determine whether an applicant
is to be selected or not, without using test
scores to tailor experiences for the individ¬
ual after he or she has been selected. As
was illustrated by the Yale study, Sternberg
has called educators’ attention to the pos¬
sibility that training after selection could be
guided by test scores indicating the student's
strengths and weaknesses. This is an ambi¬
tious, needed step forward that has been
totally ignored by most psychometrically
oriented researchers in the field.
CRITIQUE OF THE THEORY
To what extent have Sternberg and his
colleagues offered new theoretical insights
about intelligence? In order to answer this
question it is important to make a distinc¬
tion between pragmatic advances in pre¬
dicting important behaviors and theoretical
advances in the study of intelligence. Both
are valuable, but they are not the same. It
is possible to make an important pragmatic
advance in personnel selection, by calling
attention to the importance of some aspect
of intelligence already covered by other the¬
ories, but not being used for selection, with¬
out making any advance in the study of
intelligence.
In my opinion, Sternberg’s work on cre¬
ative intelligence represents a pragmatic
advance. Sternberg was not the first to dis¬
tinguish between creativity and intelligence,
nor was he the first to develop creativity
tests. Fie has made an important contribu¬
tion by calling attention to something that
was in the literature, but had “dropped off
the radar” of current test designers.
Sternberg's notion of practical intelli¬
gence is, at the theoretical level, close
to Cattell and Fiorn’s view of crystallized
intelligence (Gc). Both tests of practical
TAKING INTELLIGENCE BEYOND PSYCHOMETRICS
129
intelligence and tests of Gc stress the use
of previously acquired knowledge to solve
a current problem. Where Sternberg and
his colleagues differ from conventional test
developers is in how they have instantiated
the concept to create tests.
In conventional psychometric batteries
tests for Gc are designed to evaluate peo¬
ple's ability to use the knowledge and ideas
that are needed in industrial/post-industrial
societies, and that are often explicitly taught
in our schools. They are therefore cultur¬
ally limited, albeit to a very large, impor¬
tant culture. They are further limited by
concentrating on the evaluation of a sort
of common denominator of what people
in modern cultures are supposed to know.
This is reasonable if the test is to be
used for some broad swath of the popu¬
lation in an industrial/post-industrial soci¬
ety, such as applicants to college. Sternberg
and his colleagues have demonstrated the
feasibility of developing tests of Gc, their
"practical intelligence," for non-Western,
nonindustrial societies and for specialized
segments of the sprawling post-industrial
society. Cattell said that this would have to
be done over half a century ago.46 Stern¬
berg did it. He was not the first to do so.
Industrial-organizational psychologists have
been developing specialized tests for years.
Situational judgment testing is a good exam¬
ple. Sternberg and his colleagues have done
an effective job of showing the need for such
specialized testing in other areas.
When Sternberg and his colleagues ques¬
tion the utility of a test of, say, knowl¬
edge of English vocabulary to predict perfor¬
mance in rural Kenya, they are not attacking
the concept of Gc. They are attacking the
overgeneralization of the conventional real¬
ization of that concept, beyond the context
in which the test was developed.
All in all, Sternberg’s emphasis on prac¬
tical intelligence, if heeded, will represent
an important, albeit not unique, pragmatic
advance in testing.
The concept of tacit intelligence is a
unique theoretical contribution. A great
46 Cattell, 1957, 1971.
deal of mental competence is based on infor¬
mation and skills that we acquire without
explicit instruction. This is particularly true
of the acquisition of language and social cus¬
toms. Billions of people - literally - speak
their native language and get along in their
society without explicit knowledge of either
the rules of syntax or the rules of etiquette.
Although Sternberg and his colleagues do
not stress this, there are both input and out¬
put issues in the use of tacit intelligence.
On the input side, are some people more
able than others to pick up the unspoken
rules? Is this a general talent, or is it specific
to particular social situations? On the out¬
put side, recent research in decision making
has shown that in some situations people
are well advised to accept their gut reac¬
tion, rather than making decisions based on
conscious reflection. In other situations con¬
scious reflection is important.47 Are some
people better than others at deciding when
to act with or without reflection? These are
interesting research questions. Although the
general field of unconscious cognition has
blossomed since the beginning of the cen¬
tury, there has been little study of individ¬
ual differences in either implicit learning or
unconscious decision making.
There is no need to comment on Stern¬
berg’s analytic intelligence component, for,
as he says, this is equivalent to standard
views of general reasoning.
Sternberg has also extended his interest
to ask how intelligence is used. He distin¬
guishes three intellectual styles. By analogy
to government, he identifies an executive
style, which is an interest in putting ideas
into place; a legislative style, which is an
interest in creating ideas in the first place;
and a judicial style, which is an interest
in analyzing the implications of ideas and
efforts to put them into placed8 The exten¬
sion stresses an interaction between the “can
do" concept of intelligence and “how to"
personality and interest issues. This work
is in its beginning. The approach is an
interesting one, for the borderline between
47 Klein, 2009.
48 Sternberg, Grigorenko, & Zhang, 2008.
13°
HUMAN INTELLIGENCE
intelligence and personality research is
underdeveloped.
EVALUATION OF THE EVIDENCE
FOR THE THEORY
In Practical Intelligence (and in several arti¬
cles since that book was published) Stern¬
berg and his colleagues claim that there is
ample evidence for the three different types
of intelligence, and that practical intelli¬
gence, in particular, is an extremely impor¬
tant aspect of mental competence. I have
already expressed some reservations. Other
reviewers have expressed their reservations
with more vehemence.
Linda Gottfredson, a professor of educa¬
tion at the University of Delaware who has
frequently and ably defended classic psycho¬
metric theory, published a detailed analysis
of the evidence that Sternberg and his col¬
leagues provided in Practical Intelligence ,49
She concluded,
/The authors of Practical Intelligence]
exaggerate the strength of the empirical sup¬
port they summarize . They do so by pre¬
senting the most favorable results, over¬
stating even those, interpreting inconsistent
data in ways that produce consistent sup¬
port, and giving citations to back up strong
statements but which do not actually pro¬
vide independent support (many are just
earlier summaries of the same thing) or
that even contradict the claim in question.
The authors simultaneously discourage
the close analysis that would reveal the
inadequacies of their data and presen¬
tation. They do so partly by appealing
to many people's strong desire to believe
them, specifically by tapping the popu¬
lar preference for an egalitarian plural¬
ity of intelligences (everyone can be smart
in some way) and a distaste for being
assessed, labeled, and sorted by inscrutable
mental tests. These sentiments are evoked
again by casting aspersions on research and
researchers that have helped reinstate the
concept of g, or general intelligence.
Gottfredson, 2003a, p. 392;
emphasis in the original
Sternberg replied to Gottfredson but, as
she pointed out, did not address the spe¬
cific criticisms she made of his work.50 Stern¬
berg’s reply did make a point that Gottfred¬
son failed to address. He argued that focus¬
ing on individual studies can cause one to
fail to appreciate the broad range of sup¬
port for the theory. There is validity to this
argument.
My own evaluation is this.
Sternberg’s critiques of previous work
are too broad, especially in articles and
books written for nonspecialists. Neverthe¬
less, they have a grain of truth in them,
particularly in his criticisms of the work
of researchers who have, either implicitly
or explicitly, accepted the idea that intel¬
ligence is what the intelligence tests test.
The theoretical ideas are not as new as some
of his writing would lead one to believe,
but they do advance the field. I point espe¬
cially to his expansion of the concept of Gc
by the development of practical intelligence
tests, which could just as easily be described
as tests of the development of Gc in spe¬
cialized contexts. His ideas about possible
individual differences in the acquisition of
implicit (tacit) and explicit knowledge are
very interesting, and deserve further explo¬
ration. He is also to be applauded for attend¬
ing to the important issue of meshing assess¬
ment, training, and educational methods.
In order to support his theory he and
his colleagues have carried out a staggering
number of experiments. Taken individually,
many of these studies are fairly weak. Taken
together, they indicate that the augmenta¬
tion of conventional tests would improve
prediction. While the improvements would
not be large, they would be of practical sig¬
nificance for very large personnel selection
programs, such as admission to college or
selection of military officers.
5.3. Ackerman's PPIK Theory
Philip Ackerman, currently a professor at
the Georgia Institute of Technology, has
49 Gottfredson, 2003a.
50 Sternberg, 2003; Gottfredson, 2003b.
TAKING INTELLIGENCE BEYOND PSYCHOMETRICS
developed a theory of adult intelligence
that combines cognitive processes, person¬
ality, interests, and knowledge, hence the
PPIK model. In contrast to Gardner and
Sternberg, Ackerman has been careful to
point out that his ideas are extensions of
ideas proposed by earlier researchers. These
include R. B. Cattelfs Gf:Gc distinction and
J. L. Holland’s discussion of adult vocational
interests. Ackerman’s extensions are a sub¬
stantial and innovative contribution to our
understanding of intelligence.
Ackerman was motivated by a paradox¬
ical finding in the literature. Many studies
have shown that psychometrically defined
general intelligence (g) and, even more
telling, fluid intelligence (Gf) decline over
the adult years, beginning at about twenty-
five.51 Yet the world seems to be run by
adults over the age of thirty-five. Acker¬
man offers the following example. When
the president of Russia, Boris Yeltsin, had a
heart attack, one of the attending surgeons,
Michael De Bakey, was in his eighties. Age
was not the point; Dr. De Bakey had per¬
formed several thousand cardiac operations.
This is not an isolated example. As we
entered the twenty-first century more than
half the physicians in the United States
were over forty-five.52 Most presidents of the
United States took office when they were
in their fifties. This group includes Wash¬
ington (fifty-seven), Lincoln (fifty- two), and
Franklin Roosevelt (fifty-one), arguably the
three greatest presidents. As of 2007, 90 of
the 100 United States senators were over 50,
and 25 were over 70. Perhaps the intelligence
tests are missing something.
Following Cattell, Ackerman55 argues
that as people age they increasingly rely on
past knowledge to solve problems, rather
than solving problems by using general
problem-solving techniques. This is cer¬
tainly true for extreme cases. Studies of
expert performance in fields ranging from
chess to athletics have repeatedly shown
51 This topic is discussed further in Chapter 11.
52 Commission on Graduate Education in Medicine,
sixteenth annual report, 2005.
53 Ackerman, 1996; Ackerman & Heggested, 1997.
*3*
that very high levels of performance are
reached after a great deal of deliberate prac¬
tice, resulting in well-organized knowledge
in the domain of expertise.54 The same prin¬
ciples apply, to a lesser degree, to all adults.
Older people do worse than younger ones
on conventional intelligence tests, especially
those that stress fluid intelligence, but they
do better than younger ones when asked
to generate solutions to realistic problems,
ranging from an adolescent’s desire to leave
home to problems in credit management
and personal health.55
Ackerman was also impressed by the
extent to which adult knowledge is spe¬
cialized. Physicians, business entrepreneurs,
salespeople, mechanics, and farmers have
to meet the challenges of their particular
social niche. The situation is somewhat anal¬
ogous, although not so extreme, to the con¬
ditions that led to Sternberg’s emphasis on
culture-specific knowledge. Ackerman dis¬
tinguishes between general cultural knowl¬
edge (Gc) and knowledge of the more spe¬
cialized culture in which the individual lives
(Gk). Moreover, specialization is not lim¬
ited to professional and technical knowl¬
edge. People in our post-industrial society
will share experiences and interests with
others in groups that are broadly defined,
coherent within themselves, but different
from each other. Experienced university
professors, military officers, businessmen,
and physicians will all share some cultural
knowledge (Gc) of modern society, but they
will differ in the particular knowledge that
they have acquired within their domains
of interest (Gk). These differences have to
be considered when evaluating their intelli¬
gence.
5.3.1. PPIK Theory
Ackerman proposes that evaluations of adult
effectiveness consider four traits. The first
two of these are what we might normally
consider aspects of intelligence. Intelligence
as process refers to the ability to deal with
54 Ericsson, 1996.
55 Baltes & Smith, 1990; Baltes & Staudinger, 2000.
132
HUMAN INTELLIGENCE
novel, arbitrary, abstract problems. This
aspect of intelligence corresponds closely
to the Cattell-Horn view of fluid intelli¬
gence (Gf ), and to the mental dimension of
general reasoning (g) advocated by Jensen,
Gottfredson, and others. Intelligence as
knowledge is divided into two components:
general cultural knowledge (the Cattell-
Horn Gc, as conventionally measured) and
knowledge relevant to the individual’s inter¬
ests and general role in society (Gk), includ¬
ing but not limited to the person’s economic
role.
Ackerman stresses the development of
intelligence from the later childhood years
through adolescence and on to maturity. His
view is that over these periods intelligence is
developed by a series of investments of cur¬
rent cognitive competence into knowledge¬
building experiences that both increase and
temper the nature of future cognitive com¬
petencies. A person’s personality and inter¬
ests determine what experiences he or she
will have, and therefore will guide the devel¬
opmental process.
During childhood and adolescence com¬
petence is largely guided by the process
aspect, as the child acquires general cul¬
tural knowledge through interaction with
a relatively uniform environment, at least
in the range typical of middle-class soci¬
ety in the developed countries. The situa¬
tion is gloomier for children in extremely
poor home or school environments, which
simply do not offer much opportunity to
acquire the early stages of Gc (e.g., skill in
reading). In the later school years knowl¬
edge begins to build on knowledge, so that
Gc acquired during, for instance, the first six
years of school combines with maturing pro¬
cessing capacities (Gf) to guide acquisition
of cultural knowledge in the next six years.
This accounts for two observed phenomena,
the increasing stability of test scores as chil¬
dren progress through school and the high
validity of IQ scores obtained in early mid¬
dle school as predictors of accomplishment
by the end of high school.
Opportunities for specialized experi¬
ences arise in the later high school years,
and continue through young adulthood, in
the choice of jobs and/or college majors. By
the early to mid twenties there will be a shift
from acquisition of general knowledge of the
culture to acquisition of knowledge relevant
to specific sectors of society. Throughout
both the school and early adult years intel¬
ligence plays a crucial role, for this is what
makes the difference between having expe¬
riences and acquiring knowledge. Table 5.2
makes this point, by showing the correla¬
tions between tests of intelligence and tests
of domain-specific knowledge. Ackerman
and Beier have also demonstrated this point
experimentally. Adults were first tested for
their knowledge of the management of per¬
sonal finances, clearly an important topic
to us all. They were then given instruction,
and tested again. Both pre-test and post-test
performance and improvements in perfor¬
mance were predicted by tests of Gf and Gc.
Perhaps the smart do not just get smarter
over time; they may also get richer.
While environmental constraints un¬
doubtedly play a part (a variable that
Ackerman does not discuss), in a noncaste
society personal interests and personality
will play a substantial role in the person’s
choice of social niche, and therefore in
the subsequent development of knowledge.
This means that people who differ in how
much they know about various domains will
differ in other ways as well.
At the same time, due to biological effects
associated with aging, there will be a decline
of the processes supporting intelligence, that
is, a decline in Gf. However, this does not
mean that older people are less intelligent.
Intelligence begins to differentiate. There is
a decline in the processes supporting reason¬
ing (e.g., concentration of attention, short¬
term memory, time to retrieve information)
but an increase in the efficiency with which
those processes are used, providing that the
person is operating within a domain where
his or her previously acquired knowledge is
relevant.
Ackerman56 has identified five domains
of knowledge that are at once broader than
the knowledge associated with professional
56 Ackerman & Heggested, 1997.
TAKING INTELLIGENCE BEYOND PSYCHOMETRICS
*33
Table 5.2. Selected data on the relation between knowledge and intelligence test scores
and age. All participants were college graduates twenty-one to sixty-five years old.
Twenty-five percent had advanced degrees. Due to restriction of the range of the test
scores the estimates of correlations with intelligence are probably underestimates of the
correlations in the general population.
Content of Test
Correlation with Gf
Correlation with Gc
Correlation with Age
Chemistry
.516
.385
— .240
Physics
.568
■5*5
-.316
Electronics
•347
•451 2
.200
Law
.186
•379
.189
Geography
.320
■525
.150
Art
.168
.468
.224
World Literature
.156
.600
.248
Source: From Ackerman, 2000, Table 4. By permission of Oxford University Press and the author.
skills, yet narrower than the general cultural
knowledge evaluated by a typical test for
crystallized intelligence. The domains are
knowledge about physical science, math¬
ematics, the arts, literature, and social
science.
Knowledge in each domain is not isolated;
it is part of a trait complex that includes the
intelligence required to work in a domain
along with the personality and interests that
lead someone to decide to work in the
domain in the first place. To give these ideas
some content, imagine a longitudinal study
in which we followed students who, as they
leave high school, vary in Gc and Gf, as con¬
ventionally measured, and also vary along
the dimensions of interest in abstract investi¬
gations or social issues, and in conscientious¬
ness. Subsequently, when they are in their
thirties and forties, we test this group for
their knowledge of scientific and social top¬
ics. According to the PPIK model we would
find the following relationships:
1. Interest in abstract investigations, mea¬
sured upon graduation from high
school, would be related to adult knowl¬
edge of science. Interest in social issues,
once again measured as a high school
student, would be positively related to
knowledge of social situations.
2. Knowledge of social situations and
knowledge of science would be posi¬
tively related to general intelligence (g),
measured as an amalgam of scores on
tests of Gf and Gc. (We expect the two
to be substantially but not perfectly cor¬
related.) However, the relative weight¬
ings given to Gf and Gc might dif¬
fer depending upon whether we were
trying to predict knowledge of science
(which is often relatively abstract) or
knowledge of concrete social situations.
3. With Gf and Gc and interests held
constant, statistically, there should
be a positive relationship between
conscientiousness and knowledge.
As far as I know, this hypothetical study
has yet to be done. However, Ackerman and
his colleagues have reported cross-sectional
studies that closely resemble it.57 These
studies revealed four trait complexes, pro¬
duced by the correlations between domain
knowledge, intelligence test scores, inter¬
ests, and personality traits. They are
1. The science and math complex : This com¬
plex is defined by relatively high fluid
intelligence (g or Gf) and by interest
in realistic problems and in investiga¬
tion of abstract issues. People with high
knowledge of science and mathematics
57 Ackerman, 2000; Ackerman & Beier, 2006; Acker¬
man & Rolfhus, 1999.
*34
HUMAN INTELLIGENCE
issues tend to have high scores on this
complex.
2. The conventional complex : This complex
is defined by interest in conventional
topics, conservatism, and conscientious¬
ness. Abilities are widely distributed
across this complex.
3. The social complex : This complex is asso¬
ciated with an interest in social relations
and extraversion. Abilities are widely
distributed across this complex. Some
studies indicate that the social com¬
plex should be split into two com¬
plexes, one identifying people inter¬
ested in social power and leadership,
and the other identifying people who
value close social relations.
4. The intellectual/cultural complex: This
complex is associated with knowledge
of the arts and humanities. It is also
characterized by the personality traits
of openness to experience and a ten¬
dency to become absorbed in intellec¬
tual activities. The trait is also associ¬
ated with relatively high scores on tests
of Gc.
These descriptions are my gloss of results
from Ackerman’s studies. Slightly different
complexes, with slightly different descrip¬
tions, have been found in each study. This
is not surprising, for the studies gener¬
ally involve from one to two hundred par¬
ticipants, recruited opportunistically (e.g.,
by newspaper advertisements) rather than
by random sampling from a definable
population.
5.3.3. Evaluation of the PPIK Theory
I believe that the PPIK approach has a great
deal of potential. Ackerman and his col¬
leagues have demonstrated the need to con¬
sider domain knowledge as an attribute of
adult intelligence and the importance of
considering how personality and interests
guide the investment of current intelligence
to produce future knowledge.
I suspect that there is more to Ackerman's
concepts than the current studies have sug¬
gested. As of 2010, Ackerman's studies have
concentrated predominantly on the study of
college students and adults who have had
at least some college. It seems likely that
if studies of trait complexes were to be
extended beyond the largely middle-class,
college-educated urban population, more
trait complexes would be uncovered. Their
influence may be very powerful indeed.
5.4. Personality Variables Related
to the Development and Use
of Intelligence
The study of personality has developed
almost independently of the study of intel¬
ligence. This is a bit surprising, given the
amount of research on each topic. There is
a difference in the sense that intelligence is
a “can do” concept; it refers to the potential
for dealing with and extracting useful infor¬
mation from the environment. Personality
traits are generally “will do” concepts; they
refer to a person's disposition to take certain
actions and not others. Explanations of “can
do” and “will do” complement each other.
Psychology would benefit by an integrated
approach to the two topics.
5.4.1. Intellectual Engagement
Ackerman's ideas on the topic provide a
good place to begin our discussion. Acker¬
man has introduced the idea of intellectual
engagement . He argues that people differ
in the extent to which they engage con¬
ceptually with their experiences, especially
when such engagement is not required for
any immediate purpose. Imagine two stock
brokers, both of whom carefully check the
financial news, but only one of whom reads
the rest of the newspaper. The second bro¬
ker has more intellectual engagement than
the first.
We can go beyond thought experiments.
Goff and Ackerman developed a scale of
intellectual engagement, based upon self-
reports to questions about such things as
manner of reading a newspaper. Scores on
TAKING INTELLIGENCE BEYOND PSYCHOMETRICS
*35
this scale can contribute to the prediction
of academic performance, over a three-year
period, after the prediction due to differ¬
ences in intelligence is accounted for. 58
This research is just in its beginning. The
concept of intellectual engagement has a
great deal of promise as an explanation of
why some people learn broadly, while oth¬
ers become progressively narrower in their
thoughts as they age.
5.4.2. Self-Discipline
Intellectual engagement refers to the way
people like to behave. A related approach is
to look at attitudinal variables that deter¬
mine how much effort a person will put
into doing something he or she should do.
This brings us to the topic of self-discipline,
which I interpret as the willingness to do
something you perceive as worthwhile in
the long run, even though it means for¬
going immediate pleasures.
There is a substantial body of research
showing that the personality trait of con¬
scientiousness, that is, showing up for work
and keeping on task, is the best, and in
many cases the only, predictor of workplace
performance after the prediction associated
with general intelligence has been accounted
for. The message is clear: success in the
workplace is associated with both intelli¬
gence and willingness to work.59
The contribution of conscientiousness,
alone, as a predictor is much smaller than
the contribution of intelligence. This may
be because in many workforce situations
behavior is sufficiently constrained that con¬
scientiousness is literally forced to fall into
a narrow range. The time clock in manufac¬
turing is a prototypical example of such a
situational constraint.
The role of self-discipline seems to be
especially important in the school years.
Modern society educates its young using a
form of instruction, formal schooling, that
58 Goff & Ackerman, 1992; Chomorro-Premuzic, Furn-
ham, & Ackerman, 2006.
59 Hunt, 1995b; Schmidt & Hunter, 1998.
depends upon the student’s accepting the
idea of delayed gratification, for students are
asked to trade present fun for rather nebu¬
lous future gains. How well students meet
this challenge will determine, in part, where
they stand on the Gc scale when they grad¬
uate. Two small, but important, studies in
social psychology have shown how impor¬
tant the relationship between self-discipline
and success is.
The first study was conducted by a
research group at Columbia University.60 In
the initial phase of their study pre-school
children were offered a choice between two
rewards that varied in value. The catch was
that the children could either have the lesser
reward immediately, or they could have the
more valued one if they waited for a few
minutes. (That is an eternity to a four-year-
old.} Some fifteen years later the researchers
obtained the SAT scores that the children
had received upon their applying to college.
There was a correlation of about .4 between
the time the children had been able to wait
for the desired reward, as toddlers, and their
SAT scores as adolescents.
Duckworth and Seligman61 made the
same point in a study of self-discipline in
middle-schoolers. The children were mea¬
sured on academic achievement, intelli¬
gence, and a variety of self-discipline mea¬
sures at the beginning of the academic
year. One measure struck me as particularly
ingenious. Students were offered a choice
between $1.00 during the testing session or
a promise that the experimenter would give
them $2.00 the following week. This task
was accompanied by several other ques¬
tions in which students rated their choices
in hypothetical situations where they could
either take an immediate reward or wait
for a delayed reward. The measures of self-
discipline turned out to be better predic¬
tors for both grades and final test scores than
measures of intelligence. The relative value
is somewhat suspect, as the study partici¬
pants were students in a selective “magnet”
60 Shoda, Mischel, & Peake, 1990.
61 Duckworth & Seligman, 2005.
i36
HUMAN INTELLIGENCE
school, so there was probably a limited
range of intelligence. However, the basic
principle remains the same: self-discipline
counts.
These two studies were laboratory de¬
monstrations, using selected and probably
rather talented participants. They need to be
replicated. However, our confidence in their
conclusion is strengthened because they
complement many studies that have found
positive relations between study habits, self-
discipline, and academic performance.62
The moral to draw from this line of
research is that you will not get smart if you
do not try to get smart. But people will usu¬
ally not try something unless they think that
they have a reasonable chance of success.
This brings us to another attitudinal vari¬
able, mind-set.
Carol Dweck and her associates have con¬
ducted a large number of studies on the
relation between belief in one’s own effi¬
cacy and what people actually accomplish. 33
Dweck’s argument, which goes well beyond
the study of intelligence, is that to a consid¬
erable degree success depends upon having
a positive attitude toward chances of success
and the self-discipline to get up and act, even
when success is not guaranteed. As Henry
Ford said, “It does not matter whether you
say you can or you can't. Either way you’re
right.”64
Ford was a bit optimistic. In our soci¬
ety a person with a tested IQ in the 70s
or 80s is not a good candidate for gradu¬
ate school in mathematics, no matter how
hard he or she tries. Along with Dweck and
Ackerman, I do believe that, for any fixed
amount of intelligence, considerable varia¬
tion in accomplishment can be associated
with individual differences in self-discipline,
intellectual engagement, and in an attitude
of being willing to challenge oneself. It is
not necessary to set up a competition to
see whether intelligence or personality mea¬
sures do the best job of predicting perfor-
62 Crede & Kuncel, 2008.
63 Dweck, 2006.
64 Sign posted at the Henry Ford Museum, Dearborn,
Michigan.
mance. They are both important. Which one
is more important will depend upon the par¬
ticulars of the situation.
5.4.3. Emotional Intelligence
The study of intelligence has focused on
individual differences in people’s ability to
think, when they are thinking coolly and
rationally. We do not always act that way.
What about individual differences in our
ability to deal with emotionally charged
issues? This question brings us to the topic
of emotional intelligence.
The idea of emotional intelligence has
generated a great deal of popular discus¬
sion, although it is not always clear just what
is being discussed. Professional discussion
has revolved around three points: the extent
to which emotional intelligence ought to
be regarded as an intelligence rather than
a personality trait; the extent to which emo¬
tional intelligence is different from other,
better-researched personality and intelli¬
gence traits; and the extent to which mea¬
sures of emotional intelligence are useful
predictors of achievement in cognitive tasks,
over and above conventional measures of
intelligence.
There is a good case for the existence
of emotional intelligence.65 Intelligence is
defined as the ability to process informa¬
tion. Verbal intelligence refers to the ability
to process verbal information, visual-spatial
intelligence to the ability to process spa¬
tial and visual information, and so forth.
There is a class of information that, loosely,
can be referred to as emotional information.
This includes internal cues about one’s own
emotions and observations of the behavior
of others, which provide clues about their
emotions. Dealing with emotional informa¬
tion is as much a cognitive activity as dealing
with verbal material.
Before investigating emotional intelli¬
gence, we have to decide how it is to be mea¬
sured. By far the most common measure is
a behavioral inventory, in which people are
65 Mayer, Salovey, & Caruso, 2004; Salovey & Mayer,
1990; Salovey & Grewal, 2005.
TAKING INTELLIGENCE BEYOND PSYCHOMETRICS
*37
either asked to describe themselves or asked
how they would act in a hypothetical situ¬
ation. A second, more expensive method of
assessment is to determine the accuracy with
which people can identify emotional dis¬
plays or perform some emotion-laden task.
For example, an examinee might be asked to
identify the emotion displayed by a person
in a photograph or video display, or to list
physiological signs known to be associated
with different emotions.
Self-descriptions do not fare too well as
measurements, for they are contaminated by
respondents' tendency to exaggerate their
good qualities. This, in itself, is a confound¬
ing individual trait, for there are marked
individual differences in the extent to which
people bias their self-descriptions toward
what they perceive as socially desirable ones.
In the broad sense this is not surprising.
Shakespeare said as much some years ago.66
Even if we disregard a tendency to self-
glorification, there is no reason to believe
that people are able to accurately report
their strength in assessing and controlling
emotions, relative to other people. Doing so
would require considerable insight both into
one’s own emotional intelligence and into
the emotional intelligence of one’s acquain¬
tances. Self-reports of talents are not as
objective as test scores.67
Behavioral measures of emotional intelli¬
gence are technically better measures than
self-reports and have a firmer logical basis.
Such measures of emotional intelligence are
only modestly correlated with conventional
measures of cognitive intelligence. This is
evidence that we are indeed dealing with
two separate abilities.
Further evidence comes from the neuro¬
sciences. Following certain types of brain
injury patients display greatly reduced
ranges of emotion. In other clinical situa¬
tions the affected person will fail to dis¬
criminate emotions in others. Individuals
66 Edwards [1959) discusses the issue of social desir¬
ability. As for Shakespeare, in King Henry V the
king predicts that veterans of the battle of Agin-
court would recount “with advantage” their heroic
deeds.
67 Brody, 2004.
who have received injuries in those areas
of the brain that control emotion may dis¬
play problems in interpersonal adjustment,
but have no deficit in abstract cognitive
performance.68
There is no contention that the brain
mechanisms for dealing with “hot” and
"cold” cognition are entirely separate. All
that needs to be shown is that they are some¬
what different. We now know that the per¬
ception and activation of emotion are both
associated with the limbic system, especially
the amygdala, and the ventromedial pre¬
frontal cortex.69 This contrasts with find¬
ings on the neural correlates of intelligence,
which indicate heavy involvement of the lat¬
eral prefrontal cortex and the anterior pari¬
etal cortex in cognitive activities.70
The combination of the neuroscientific
and correlational evidence makes a con¬
vincing case that there is such a thing as
emotional intelligence, and that it is not
the same as cognitive intelligence, although
there is a small correlation between the
two. A practical question remains. Does
emotional intelligence predict academic and
workplace performance as well, better, or in
addition to measures of general (cognitive)
intelligence?
Here there seems to be a great deal of
“sound and fury” but, at this time (2010),
rather little reliable evidence.
In his popular book on the topic
Goleman71 asserted that emotional intelli¬
gence is a much more important determi¬
nant of behavior than cognitive intelligence.
He made his case largely by anecdote and
informal reports. There are a number of
studies showing that emotional intelligence,
if measured alone, correlates with desir¬
able outcomes in life. The key question,
though, is whether measures of emotional
intelligence can predict important behav¬
iors, beyond those that can be predicted by
conventional intelligence and personality mea¬
sures. Few such studies have been reported,
68 Bar-On et al., 2003.
69 Damasio, 1994, 1998.
70 Jung & Haier, 2007.
71 Goleman, 1995.
138
HUMAN INTELLIGENCE
and those that have appeared have received
sharp criticism.72
Despite the present record, though, I
would not dismiss the potential contribu¬
tion that emotional intelligence may make
to the study of intelligence. I suggest that
these contributions will be greatest when
we examine “will do” rather than “can do”
performance, over a fairly long period of
time, and when the performance being eval¬
uated requires interpersonal interaction, in
face-to-face situations. A salesperson taking
orders in a telephone call center for a ready¬
made clothing store does not have to have
high emotional intelligence; a salesperson in
an expensive clothing boutique does.
5.5. Closing Comments on Models
to Expand Beyond Psychometric
Intelligence
It is unlikely that dissatisfaction with con¬
ventional tests and testing will go away.
While the critics have made some good
points about the shortcomings of modern
testing, they have not come up with good
alternatives. If we are going to develop a sci¬
entific understanding of any topic, including
human intelligence, there has to be some
way of measuring what we are studying. Any
claim that intelligence, as currently mea¬
sured, is not important simply flies in the
face of the facts. However, progress depends
upon going beyond the known.
Consider another of Gardner's com¬
plaints about modern psychometric theory:
Frankly I find it difficult to find a publicly -
discussable reason why researchers should
devote their energies to yet one more study
of the role of *g' in arenas ranging from job
performance to longevity or yet one more
effort to document racial or ethnic differ¬
ences in intelligence.
Gardner , 2006, p. 300
This is a pugnaciously worded claim that
psychometric research is spinning its wheels.
While I would not be as strong as Gardner
72 Brody, 2004; Matthews, Zeidner, & Roberts, 2005.
in making such a complaint, a great deal of
current work is highly repetitive. The field
could use a paradigm shift. I do not see
how this can come about without the devel¬
opment of new methods of measurement.
Where are we in this endeavor?
Instead of standing on the faces of those
who went before, as psychologists are wont
to do, let us stand on their shoulders. Any
new theory of intelligence will have to deal
with the facts established by present psy¬
chometric findings. General intelligence, g
as measured by conventional testing, exists,
and is a trait associated with accomplish¬
ment (or lack of it) in virtually all aspects of
our society, not just in scholastic endeavors.
There are two acceptable models for the
psychometric data: the three-stratum model
and the g-VPR model, both described in
the previous chapter. New models should
build on the insights contained in these
two approaches. Which one a new theorist
should choose to emphasize depends to a
great extent on what that theorist wants to
accomplish.
If the goal of the new theory is to explain
how intelligence functions in society, as
Ackerman and his colleagues wish to do,
the three-stratum theory is a good starting
point. If the goal is to probe more deeply
into the processes of intelligence, especially
to the point of establishing the biological
basis for individual differences in cognitive
power, the g-VPR model may be the best
starting point. I elaborate on this idea in the
next chapter. In either case there has to be
a starting point.
Attempting to develop theory without
considering history is simply not a good idea.
Of the three “new intelligences” reviewed
here I believe that Gardner has disregarded
the historical record, to the considerable
detriment of his ideas. Sternberg has built
on previous work. Two of his ideas, the
need to study tacit knowledge and the need
to incorporate research on creativity into
research on intelligence, are clearly a step
forward. Ackerman and his colleagues have
built on previous research findings. Their
work points a way toward a better under¬
standing of intelligence in everyday life.
TAKING INTELLIGENCE BEYOND PSYCHOMETRICS
139
While these new theorists have not pro¬
duced any revolutionary new paradigms,
they have outlined an agenda for the future.
It turns out that it can be summarized
by an e-mail message that I received just
as I finished writing this chapter. The
message, which was intended as a joke, con¬
trasted examinations in engineering and psy¬
chology.
The question for psychology was:
Based on your knowledge of their works ,
evaluate the emotional stability, degree
of adjustment, and repressed frustrations
of the following: Alexander of Aphro -
disias, Ramses II, Gregory of Nicia,
Hammurabi. Support your evaluation
with quotations from each man 's work,
making appropriate references. It is not
necessary to translate.
The question for engineering was:
The disassembled parts of a high-
powered rifle have been placed on your
desk. You will also find an instruction
manual, printed in Swahili. In ten
minutes a hungry Bengal tiger will enter
the room. Take whatever action you feel
appropriate . Be prepared to justify your
decision.
Leaving aside the humor, these questions
illustrate some important principles for the
design of new assessment methods.
The question about psychology is humor¬
ous because answering it requires retrieving
and rearranging knowledge. The task itself,
writing a report, is a natural one. Doing so
requires the exercise of important cognitive
skills that, by their nature, take time to
execute. Such skills cannot be evaluated
within the testing paradigm. Nevertheless,
they are very important aspects of cognitive
competence.
What about the engineering question?
Given proper training, assembling a rifle is
not difficult. Soldiers and marines in mod¬
ern armies all learn to do this. During learn¬
ing the trick is to be able to visualize how
the various parts should be moved to fit
together. This is an exercise in visual rota¬
tion, the R dimension in the g-VPR model
and the second-order Gv trait in the three-
stratum model.
What we have here is a test of domain
knowledge, Gk in Ackerman’s expositions,
and of visual-spatial skills.
The problem is the tiger, not the rifle.
The task must be done in a hurry ( upesi in
Swahili) . Interfering thoughts of tigers must
be suppressed. The examinee has to manage
his or her emotions in a way that facilitates
the cognitive task rather than paralyzing it.
A similar problem occurs during knowl¬
edge acquisition. Military boot camp, where
you learn to assemble a rifle, is (intention¬
ally) an emotional experience. Learning is
possible only if motivation can be main¬
tained and emotions can be suppressed.
In boot camp the emotions to be sup¬
pressed may be fear and confusion. In school
and the workplace different emotions are
involved (boredom? self-consciousness?),
but the point is the same. Understanding
the development of intelligence as knowl¬
edge requires an understanding of how intel¬
ligence interacts with motivation, personal¬
ity, and situational constraints to produce
understanding .
The present attempts to expand intel¬
ligence only scratch the surface of these
issues. Too much effort has been spent try¬
ing to claim revolutionary new advances or,
even more destructively, setting up opposi¬
tions between cognitive abilities, personality
characteristics, and environmental variables.
Past cognitive abilities, personality charac¬
teristics, and interests interact to increase
intelligence. Progress depends on our devel¬
oping an agenda for research that will
describe the interaction.
CHAPTER 6
The Mechanics of Intelligence
I believe I have an unfair edge over most of
my colleagues right now - my mind works
better than my mouth does.
U.S. Senator Tim Johnson, from
a speech in Sioux Falls, South
Dakota, announcing his intention
to return to the Senate following a
brain hemorrhage that left his
speech impaired, August 28, 2007
Psychometric theories use what Sternberg
has called a geometric analogy for intel¬
ligence; people are seen as varying along
dimensions of intelligence in much the same
way that they vary along the dimensions of
height and weight.1 Different theories iden¬
tify different dimensions, but the geometric
analogy is maintained. This is a useful way of
summarizing variations in intelligence across
populations, but it has a serious shortcom¬
ing. The geometric analogy does not explain
the processes that make up thinking.
To see what this means, imagine two
individuals, Ignatz and Horatio. We first
1 Sternberg, 1990.
determine their psychometric intelligence,
in terms of the g-VPR model, and then ask
them to attack the following two problems.
The first problem makes use of the English
rules that permit center embedding, putting
one relative clause inside another. Ignatz and
Horatio are presented with sentences of the
form
The rat ate the cheese
The rat the cat chased ate the cheese
The rat the cat the dog scared chased ate
the cheese
The rat the cat the dog the man owned
scared chased ate the cheese.
And so on
and we determine at what point each person
finds the sentence incomprehensible.
The second problem is one of those
(in)famous mathematics word problems:
The distance between Seattle and San
Francisco is approximately 750 miles by
air. If an aircraft leaves San Francisco at
9 a.m. and travels toward Seattle at 500
miles per hour at an altitude of 37,000 feet ,
while at the same time an aircraft leaves
Seattle and travels to San Francisco at
140
THE MECHANICS OF INTELLIGENCE
M1
55 o miles per hour at an altitude of 34,000
feet, when will they pass by each other, and
how far is this point from Seattle?
To complete the example, assume that
we had conducted an experiment, not
involving Ignatz and Horatio, in which prob¬
lems like these were included in a large
psychometric study using the tests covered
by the g-VPR model. If we knew Ignatz’s
and Horatio’s scores on the g-VPR dimen¬
sions, we could determine the probability
that each of them would correctly solve the
two problems. We could not go beyond
this because the psychometric approach
does not provide models of the problem¬
solving process. Psychometrics does not tell
us what the elementary steps are in the
problem-solving process, or how individual
differences in the ability to execute these
steps translate into individual differences in
problem-solving ability, including the abil¬
ity to take intelligence tests. To go further
we have to look at another branch of psy¬
chology: cognition.
6.1. The Cognitive Psychology
Approach
Suppose that we were to ask a modern cog¬
nitive psychologist to explain Ignatz’s and
Horatio’s problem-solving behavior. The
cognitive psychologist would want to look
at the requirements of the problem-solving
task. The psychologist would first observe
that Ignatz and Horatio would have to be
able to retrieve word meanings. This, in
itself, is not a trivial task. Then there is more.
In order to untangle center-embedded sen¬
tences a person has to have some way of stor¬
ing information about a noun phrase until it
can be connected to its verb. To understand
The rat the cat chased ate the cheese
Ignatz and Horatio would have to have some
way of temporarily storing the phrase the
rat, processing the cat chased, attaching the
result of the processing to the rat (i.e.,
the particular rat that the cat chased], and
connecting the modified idea to the verb
phrase ate the cheese. The cognitive psychol¬
ogist would want to know what temporary
storage capacities Ignatz and Horatio had
for holding unresolved noun phrases, and
whether they could organize that storage so
that the noun phrases would be available
when their verbs were encountered.
Turning to the mathematics word prob¬
lem, the cognitive psychologist would first
observe that Ignatz and Horatio have to
retrieve word meanings, and have sufficient
temporary storage space in memory so that
they can analyze the meanings of the sen¬
tences. However, the passage does not con¬
tain deeply center-embedded sentences, so
sentence comprehension itself would not
present a challenge to their capacity for tem¬
porary memory storage. There would be
another challenge.
Using information extracted from the
sentences, Ignatz and Horatio would have
to develop an internal representation of the
problem. This could either be in the form
of equations or in the form of a ‘picture in
the head.” The cognitive psychologist would
probably remark that both representations
require internal storage of information, but
differ in the kinds of information stored. A
representation in terms of equations is a
symbolic representation that at least approx¬
imates syntactical analysis. A maplike rep¬
resentation requires a visual-spatial mem¬
ory store. The cognitive psychologist would
also note that the problem requires answer¬
ing two questions, the time the two aircraft
meet and the distance from Seattle. The
problem solver has to set up a goal-subgoal
structure to decide which problem to attack
first, work on that problem, and then switch
to the second problem. The cognitive psy¬
chologist would like to have an estimate of
how many goals and subgoals Ignatz and
Horatio can keep track of, and how long it
takes them to switch from working on one
goal to working on another.
Stepping back from the particulars of the
two problems, cognitive psychologists offer
explanations of behavior in terms of elemen¬
tary cognitive tasks (ECTs), such as retrieving
the meaning of a letter or word, holding a
piece of information in temporary storage,
142
HUMAN INTELLIGENCE
and switching attention from one cogni¬
tive task to another. Information-processing
theories of intelligence attempt to relate
individual differences in intelligence, in the
broad sense of the ability to function in the
world, to individual differences in the exe¬
cution of elementary cognitive tasks.
The cognitive psychologist stresses that
such explanations are bound to be incom¬
plete, because problem-solving ability will
be determined both by the information the
person has (knowledge) and how well he
or she can manipulate information, in the
abstract. An information-processing model
addresses only the second question. Incom¬
plete as it is, it is an important question
because information-processing capacities
stand somewhere between psychometric
models of intelligence and brain-based
explanations of intelligence.
To see this, return to the example of the
center-embedded sentence. Knowing where
Ignatz stands in the geographic space defined
by the g-VPR model tells us the probability
that he will be able to understand a sentence
with, say, three levels of embedding. Spec¬
ulating a bit, based on the discussion of the
brain to be presented in the next chapter,
we might observe that when Ignatz tries to
understand a center-embedded sentence the
frontal, anterior parietal, and left temporal
parts of his brain show heightened activity.
Horatio might be located at a different point
in the g-VPR space and have a different
probability of understanding the sentence,
and while trying to understand the sentence
Horatio might show different levels of acti¬
vation than Ignatz does. It is useful, even
necessary, to have an explanation that stands
somewhere between the psychometric and
brain-based explanations, by explaining in
functional terms how Ignatz and Horatio
are limited by their information processing
abilities.
6.1.1. An Historical Note
The idea that individual differences in cog¬
nitive power are partially due to underly¬
ing differences in the ability to perform ele¬
mentary cognitive tasks is not new. In the
mid nineteenth century Galton measured
the speed with which people could per¬
form simple tasks, such as making a sim¬
ple movement in response to an auditory or
visual signal, and attempted to find a cor¬
relation between reaction time and occupa¬
tional status. The results were not impres¬
sive, so he apparently dropped this line of
investigation. In 1901 Clark Wissler, a grad¬
uate student working under the supervision
of James McK. Cattell, attempted to find a
relation between tasks similar to those Gal¬
ton had used and the grades of Columbia
University students. This work was also con¬
sidered to be unsuccessful, and for many
years the “revealed wisdom” was that there
is no correlation between intelligence and
performance on measures of information
processing.2
In retrospect the conclusion was pre¬
mature. Galton and Wissler both expected
to find very strong relationships between
performance on elementary cognitive tasks
(ECTs) and complex thought. Today we
know that the relations are there, but
that they are moderate. Both Galton 's and
Wissler’s techniques for measuring reaction
times were not close to modern standards.
Probably the biggest mistake they made was
that they did not take into consideration
variation in the execution of an ECT within
the individual. They would, for instance,
determine a person’s reaction time by aver¬
aging three or four attempts to execute
the task, whereas today upward of a dozen
measurements would be taken. (Even today
cognitive psychologists sometimes criticize
researchers interested in intelligence for bas¬
ing their measurements on too few trials.)
Wissler’s study had low statistical power,
which means that it was capable of detecting
only large relationships. All in all, by today’s
standards the research was so poorly done as
to be virtually useless.3
The deficiencies in the research were not
appreciated at the time, so the perceived
failure of the Galton-Wissler approach
resulted in a virtual cessation of interest
2 Jensen, 2006, pp. 5-7.
3 Deary, 1994.
THE MECHANICS OF INTELLIGENCE
in individual differences in elementary cog¬
nitive tasks. Psychometric studies of intel¬
ligence flourished, and, especially after
the mid-1950s, the study of information¬
processing models, under the title cognitive
psychology , became the dominant method
for studying human thought, without regard
for individual differences. These fields were
pursued as separate endeavors, however,
involving different investigators and pub¬
lication in different journals. There were
a few calls for a rapprochement, notably
in a paper by the educational psychologist
Lee Cronbach in 1957.4 However, very little
was actually done until 1973, when my col¬
leagues Nancy Frost and Clifford Lunneborg
and I published a series of studies on the
information-processing correlates of verbal
and mathematical reasoning.5 This work was
followed by a blizzard of studies, certainly
not limited to my laboratory.
6.1.2. Modern Information-Processing
Models
In order to understand the attempts to
connect information-processing models of
thought to intelligence we must understand
modern theories of cognition as information
processing. These theories owe a great deal
to concepts developed for designing digital
computing systems. However, the modern
view in no way implies that the digital com¬
puter can be used as a model of the mind.
Rather, the contention is that any compu¬
tation, including thought, has to rely on
certain elementary actions, and that study¬
ing performance on the individual actions
is a useful way to proceed when studying
cognition.
Any computing device much more com¬
plicated than a light switch has to be
able to do three things: sense the envi¬
ronment (perception], classify the environ¬
ment into states relevant to the device (cat¬
egorization], and relate these classifications
to previously stored information (memory
retrieval]. The result of these computations
4 Cronbach, 1957.
5 Hunt, Frost, & Lunneborg, 1973.
H3
is an internal representation of the current sit¬
uation as interpreted in the light of memory
of past situations. The internal representa¬
tion may then be used to select a response
(decision making]. If the computing device
operates continuously, in a dynamic, chang¬
ing environment, it has to be able to main¬
tain a record of the situations it has encoun¬
tered (as interpreted], the responses that
have been made, and the results of the
responses. Therefore, we have to be con¬
cerned with the storage and retrieval of
information in long-term memory, in a way
that ensures its accessibility when needed.
Although one can conceive of a robot
mind that executed each of the four tasks,
perception through decision making, in
serial order, in the human brain they are
interleaved, with a great deal of feed¬
back between them. A model of how this
exchange of information takes place is called
a cognitive architecture . Psychologists who
study human information processing have
converged on some version of the Black¬
board model of cognitive architecture.6 A
version of this model is shown in Figure 6.1.
The components of the model are as follows:
1. Information from the environment is
sensed, and then classified into progres¬
sively higher-order categories, through
arousal of related information in long¬
term memory. Long-term memory is
not thought of as a static storage pro¬
cess, as is the case in computer systems.
Items in long-term memory exist in vari¬
ous states of activation, depending upon
how frequently and how recently they
and related pieces of information have
been attended to. Therefore, the inter¬
pretation of a percept will be influenced
by the context in which it occurs.
2. The information from the environ¬
ment, together with relevant informa¬
tion retrieved from long-term memory,
is placed in working memory, a term
6 Anderson, 1996; Hunt & Lansman, 1986; Meyer &
Kieras, 1997; Newell, 1990. These references sum¬
marize the ideas that the various authors had put
forward over a number of years. I certainly do not
claim priority.
i44
HUMAN INTELLIGENCE
Figure 6.1. The blackboard model of cognition. Information about
the current situation is held in a temporary working memory store.
Sensory input enters working memory if attention is being directed
to the appropriate input channel. Information on the blackboard and
from the sensory input channel activates problem-solving procedures
that have been stored in long-term memory. These place further
information in working memory. Sensory information may also enter
long-term memory directly, thus lowering the threshold required to
activate some problem-solving procedures or, in some cases,
initiating an action without placing information on the
blackboard.
introduced by the British psychologist
Alan Baddeley,7 and widely accepted
today. There further processing results
in an internal representation of the cur¬
rent situation, as interpreted. The
internal representation may include
interpretations in contexts, and the
identification of goals and subgoals in
problem-solving. In totality, the pro¬
cesses acting on the internal representa¬
tion are referred to as executive processes ,
whose role is to update the internal
representation and establish priorities
for action. One of the most important
subsets of the executive processes are
the attentional processes that highlight
some pieces of information and suppress
others.
3. Organizing information in this way
enables a person to appear to be deal¬
ing with two or more tasks simulta¬
neously - for example, talking while
driving an automobile. The impres¬
sion of simultaneity is something of
an illusion, as both tasks will compete
7 Baddeley, 1986.
for space in working memory and for
attention. When two tasks are done
together, therefore, close examination
almost always shows that one or both
of the tasks are performed less well than
they would be if performed alone.
4. Working memory acts as a “blackboard”
that broadcasts information into long¬
term memory. Alternatively, you can
think of long-term memory processes as
actively watching working memory to
see if their cue has been called, rather
like actors at a stage production.
5. The storage section of working mem¬
ory is thought of as being divided into
modality-specific sections for linguis¬
tic, auditory-nonlinguistic, and spatial/
visual information. There is no implica¬
tion that working memory is located at
a single place in the brain. It is a system
resulting from the integration of pro¬
cessing at several different locations.
6. Information in working memory, as
interpreted, is consolidated in long-term
memory. This takes time. Therefore,
the, probability that a piece of informa¬
tion in working memory will be stored
THE MECHANICS OF INTELLIGENCE
in long-term memory is partially deter¬
mined by the time it remains in working
memory. Items briefly attended to are
not likely to be remembered.
7. From time to time the information in
working memory will initiate a process
that calls for a motor response - speak¬
ing or making a physical movement.
While many information-processing
psychologists have been quite con¬
cerned with response production, 8 psy¬
chologists interested in intelligence have
done rather little work on this aspect of
cognition.
Cognitive psychologists are concerned
with the typical characteristics of each of
the processes in the cognitive architecture.
Psychologists interested in intelligence want
to know how individual differences in these
characteristics are related to individual dif¬
ferences in intelligence, as measured by psy¬
chometric models. Using the g-VPR model
as a guide, we will first look at two aspects
of the architecture that appear to be related
to individual differences in general rea¬
soning, and then examine specialized pro¬
cesses related to verbal and visual- spatial
reasoning.
6.2. The Speed of Mental Processing
The blackboard model depicts thinking as
the shuttling of attention from one piece of
information to another. To illustrate this,
consider the common task of determining
whether a view of a scene matches a descrip¬
tion. The task could be as simple as answer¬
ing the question Are my car keys on the dining
room table? The listener must form an inter¬
nal representation of the linguistic state¬
ment, form an internal representation of a
visual scene, and compare the two. Different
people use different strategies for comparing
sentences to pictures,9 but no matter which
strategy they use, they have to shift back and
8 For instance, Meyer & Kieras, 1997.
9 MacLeod, Hunt, & Mathews, 1978.
M5
forth from processing one piece of informa¬
tion to another.
All this has to happen quickly enough
that the thinker can keep up with the envi¬
ronment. If a speaker is speaking at a normal
conversational tempo, listeners are expected
to keep up. See the example and elaboration
in panel 6.1.
There is also an internal demand for speed
in mental processing. Competing lines of
thought vie for the limited storage space
in working memory, and the competition
is fierce. Have you ever been introduced to
someone, immediately embarked on a con¬
versation, and then realized that you did
not know the person’s name, even though
the two of you had just been introduced? If
information in working memory is usurped
by another task, the original contents of
working memory may be lost before it can
be moved to long-term memory.
These considerations show why speed of
processing should be important in thought.
But is there a single speed of processing, or
different speeds for different ECTs?
The reason for believing in processing
speed as a pervasive trait is that all men¬
tal actions depend on neural processing. If
people differ in the effectiveness of pro¬
cessing speed at the neural level, then there
should be pervasive individual differences in
processing speed no matter what the task.10
Accordingly, speed of processing should be
positively correlated with measures of g, the
general reasoning factor. In order to test this
hypothesis we have to find a way of measur¬
ing mental processing speed.
6.2.1. Donders’s Paradigm as a Way of
Measuring the Speed of Mental Processing
Elementary cognitive processes cannot be
measured in isolation. Even the simplest
tasks will require multiple mental steps, and
may mix cognitive and noncognitive pro¬
cesses. In the nineteenth century Galton
asked people to strike a bag upon a sig¬
nal. He took the time interval between
10 Jensen (2006] has developed this argument at con¬
siderable length.
146
HUMAN INTELLIGENCE
Panel 6.1. Intelligence and the Speed
of Language Comprehension
The speed with which people can com¬
prehend simple speech is often taken as
an indicator of intelligence. I will illus¬
trate with an anecdote and an actual
demonstration.
The anecdote is based on an inci¬
dent in my laboratory. Most of my lab¬
oratory research effort at that time was
devoted to studying individual differ¬
ences in cognition in young adults, usu¬
ally college students. However, one grad¬
uate student was interested in mentally
disabled children. In order to communi¬
cate with the participants in her studies
she learned to . . . talk . . . very . . . slowly.
When she reported her results to her fel¬
low graduate students she . . . drove . . .
them . . . up . . . the. . .wall.
Recall Baddeley’s three-minute test of
reasoning, described in Chapter 2.* Peo¬
ple were shown a picture and a state¬
ment, and asked if the picture described
the statement. Two examples, one easy
and one hard, are
Is the A after the B ? AB
Is the B not before the A? BA
The score on this test was the num¬
ber of items that could be answered in
three minutes. It had a .59 correlation
with scores on a formal intelligence test
used to screen recruits in the British
army. Tasks that require coordination
of linguistic statements and visual infor¬
mation emerge as a way of evalua¬
ting g.
* Baddeley, 1968.
the signal and contact with the bag as a
measure of speed of processing. A mod¬
ern cognitive psychologist would point out
that this reasoning is flawed. The interval
between the signal and the strike contains
many subprocesses: detecting that a signal
has been received, interpreting it, selecting
a response, and making the movement. The
processes of detecting and interpreting the
signal and constructing the response are cog¬
nitive acts, but hand movement is at best
semicognitive, as the time taken to move
the hand would depend on properties of the
muscular-skeletal system as well as the ner¬
vous system. Someone with arthritis would
strike a bag very slowly, but arthritis does
not affect cognition. Some way has to be
found to isolate each of the several cognitive
actions that lie behind an overt response.
Franciscus Donders, a nineteenth-
century Dutch physiologist, proposed a
solution to this problem. Suppose that a
task can be fractionated into a sequence of k
steps that must be conducted in sequence.
Suppose further that you can construct a
second task that requires all the steps of
the first task, and an additional k + ist step
that has to be inserted somewhere in the
series of steps required by the first task. The
difference between the time required to
execute the first task and the time required
to execute the second task will be a
measure of the time required to complete
the inserted step. That is, if we let R(x) be
the time required to complete process x,
then
Reinserted process) = R(task with insertion)
- R (original task). (6.1]
This brief explanation does not really do
justice to Donders’s sophisticated ideas, but
it is close enough for present purposes.
Donders’s approach is based on three
assumptions: that the processes involved
must be executed in sequence, that for any
pair of adjacent processes in the sequence
the first process is not initiated until the sec¬
ond is completed, and that the speed with
which a process is completed is independent
of the speed with which any other process
is completed. Collectively these conditions
THE MECHANICS OF INTELLIGENCE
H7
are referred to as the independent serial pro¬
cessing assumptions. The literature in cogni¬
tive psychology contains many discussions of
how one can ensure that independent serial
processing has occurred, for the issue is sub¬
tler than you might think.
Donders’s logic applies only to situations
in which the entire task is completed suc¬
cessfully. The effect of an error in any one
of the processes is not defined. Therefore, a
careful investigator should restrict his or her
analyses to trials on which a correct response
is made.
Unfortunately, researchers interested in
intelligence have not always paid as much
attention as they should either to the need
to prove that the independent serial pro¬
cessing assumptions apply to their tasks, or
to restricting their analyses to correct tri¬
als. In spite of these limitations, Donders’s
paradigm has proven to be useful. Panel 6.2
provides two examples of how Donders’s
method of subtraction has been used to mea¬
sure the time required to access information
in long-term memory. Panel 6.3 provides an
example of how it can be used to measure
the time needed to access information in
working memory.
Experimental paradigms have also been
designed to measure the speed with which
a person can make an elementary decision,
without reference to memory. Such studies
are called choice reaction time (CRT) tasks.
At the beginning of the experiment the par¬
ticipant is told that one of a small number of
directly perceivable stimuli may occur. For
instance, the participant might be told that
either a red or a green light could be pre¬
sented. The participant’s task is to identify,
as quickly as possible, which of the possible
stimuli has actually been presented.
Choice reaction tasks differ in the num¬
ber of possible stimuli that might be pre¬
sented. Discriminating between red and
green lights is an example of a two-choice
reaction time task , denoted 2CRT. In a
more complicated experiment, if the choice
were to be among red, green, yellow, and
blue lights, the task would be a 4CRT task.
In practice experiments seldom go beyond
the 8CRT task.
The 2CRT condition represents the time
required to make a binary decision. On log¬
ical grounds, this should be central to cog¬
nition. However, the response time in the
2CRT condition will also include the time
required to make the response. This contam¬
inates cognitive and noncognitive processes.
Suppose that an examinee had broken a fin¬
ger prior to the experiment. Such a person
would be slow pressing a button, and hence
have a long reaction time. It would hardly be
fair to infer that someone’s brain was work¬
ing slowly because of a broken finger.
To avoid such problem we can measure
the manner in which choice reaction times
increase with the number of choices. One
such analysis is described in panel 6.4. As
the panel explains, an analysis of how reac¬
tion time changes as the number of choices is
increased can provide an estimate of cogni¬
tive processing speed, independent of motor
processes.
6.2.2. Identification Paradigms
Donders's approach depends upon the
assumptions of strict serial processing and
independence of processes from each other.
An alternative approach is to construct a task
that is believed to be primarily responsive to
a particular cognitive process, and use it as
a measure of the time required for that pro¬
cess, ignoring the fact that other processes
may also influence the measure.
One such paradigm is the lexical identifi¬
cation task, which is an attempt to measure
the time required to retrieve information
from long-term memory. An experimen¬
tal participant decides that a letter string
either does or does not constitute a common
word, for example, MALEC or CAMEL.
Reaction time (RT) is measured. A gener¬
alization of this task is the semantic identi¬
fication task. In semantic identification an
object name and a category name are pre¬
sented, for example, CAMEL ANIMAL.
The participant indicates whether or not
the object named is a member of the cate¬
gory. In such tasks the reaction time reflects
both retrieval of information from long-term
memory and the motor processes involved
148
HUMAN INTELLIGENCE
Panel 6.2. Using Donders's Logic
to Measure Access to Long-term
Memory
The following two examples show how
Donders’s paradigm can be used to mea¬
sure the speed of access to long-term
memory.
Example 1. Category identification
Task 1. The examinee sits in front of
a computer monitor. A program presents
single digits on a screen. The examinee is
to depress the space bar as soon as a digit
is presented.
Task 2. The procedure is identical,
except that the space bar is to be
depressed only if an odd digit is pre¬
sented.
The first task can be conceptualized
as
(1) Number presented. (2) Recognize
number. (3) Press space bar. Let the time
required to do this be RT(i).
The second task can be conceptualized
as
(1) Number presented. (2] Recognize
number. (3) Classify number and select
response. (4) Press space bar if appropri¬
ate. Let the time required to do this be
RT(2).
Applying Donders’s logic, the time
required to make a simple decision is
Time to classify stimulus and select
response = RT(2] — RT(i}.
Example 2. Name identification and
physical identification
This task was originally developed by
Michael Posner and his colleagues at the
University of Oregon.* Variants of it have
been used widely in studies in cognitive
psychology.
The examinee is shown two letters on
a computer screen and is asked to indi¬
cate, as rapidly as possible, whether or
not the two letters are identical. Possible
letter pairs are
A A
A B
A a.
The answer for these three pairs
depends upon what is meant by “iden¬
tical.” Does this mean that they are phys¬
ically identical, or does it mean that they
name the same letter? For the first pair the
answer is “yes” in each case, for physi¬
cally identical objects must have the same
name. For the second pair the answer is
“no.” For the third pair, A a, the answer
depends upon the instructions; the let¬
ters are physically different but refer to
the same letter of the alphabet.
Now suppose that identification takes
place in the following steps.
Physically identical (PI) steps
Pi. Determine whether the two letters
are physically identical.
P2. If they are, respond “yes”; otherwise
respond “no.”
Name identical (NI) steps
Ni. Determine whether the two letters
are physically identical.
N2. If they are, respond “yes”; otherwise
go to step N3.
N3. Retrieve letter names.
Nq. Determine whether the two letters
have the same name.
N5. If they do, respond “yes”; otherwise
respond “no.”
In order to perform under name-
identical instructions, the observer has
to retrieve the names of letter forms,
an act involving retrieval of information
from long-term memory. This takes time.
College students take about eighty mil¬
liseconds longer to execute the identi¬
fication task under the NI than under
the PI condition. Thus eighty millisec¬
onds can be taken as a rough estimate
THE MECHANICS OF INTELLIGENCE
149
of the time required to access highly
overlearned information in long-term
memory.
The eighty milliseconds is the average
identification time taken by college stu¬
dents in general. If college students are
split into two groups, one with high ver¬
bal comprehension test scores and one
with low test scores, students in the high
test score group take about sixty mil¬
liseconds to deal with the NI condition,
compared to the Cl condition, and stu¬
dents in the low test score group take
about ninety milliseconds.1
* Posner et at, 1969.
Hunt, Frost, & Lunneborg, 1973; Hunt,
Lunneborg, & Lewis, 1975.
Panel 6.3. The Short-term Memory
Scanning Task
The short-term memory scanning task
was developed in the 1960s by Saul
Sternberg, then a researcher at the
(now disbanded] Bell Laboratories of
the American Telephone and Telegraph
Company.* Since then the paradigm has
been widely used by many researchers.
The procedure is as follows:
A memory set of from one to six digits
or letters is presented - for example,
A Q X L.
The memory set is removed and a
probe item is presented - for exam-
pie, T.
The examinee indicates, by pressing a
button, whether or not the probe was
a member of the memory set. In this
case the answer is NO.
The procedure is then repeated, with
a new memory set.
The scoring procedure illustrates the
use of mathematical modeling. Let k be
the number of items in the memory set,
and let RT be the reaction time, that is,
the time between the onset of the probe
item and the point at which the exami¬
nee makes a response. We consider only
correct trials. It is considered important
to train participants until they make very
few errors. Otherwise, individual differ¬
ences in trading off speed and accuracy
will influence the results. Unfortunately,
this caution has not always been followed
in research on intelligence. Empirically
it has been found that after individuals
have received enough training that they
make errors on only about 5% of the trials,
RT increases linearly with the number of
items in the memory set:
RT — A -\- Bk,
where A and B are positive constants.
The B parameter can be thought of as the
increment in time required to scan the
memory set when memory set size is
increased by adding one item. This inter¬
pretation makes it a reasonable measure
of the speed of access to information in
short-term memory. The A parameter,
by contrast, reflects all of the processes
involved, including identifying the probe
item and making the response.
Sternberg’s analysis of memory scan¬
ning is a variant of Donders's method,
because the estimate of the B parameter
is sensitive to the successive differences
in the times required for memory sets of
one versus two, two versus three, and so
forth.
* S. Sternberg, 1966.
i5o
HUMAN INTELLIGENCE
Panel 6.4. Jensen’s Choice Reaction
Time Procedure
Arthur Jensen developed a choice reac¬
tion time procedure that is widely used in
the study of individual differences.* The
examinee is seated in front of a display
that consists of (a) a bank of eight lights,
with buttons immediately below them,
and (b) a “home” button in the center of
the display. Depending on the condition,
the examinee is told that any one of two,
four, six, or eight lights may be lit. When
one of the lights comes on, the examinee
lifts his or her finger from the home but¬
ton and depresses the button under the
light that has been lit.
This procedure yields two measures
on each trial. The decision time is the
interval between the time the light is
illuminated and the time the examinee
releases the home button. The movement
time is the interval between the release of
the home button and the depression of
the button under the illuminated lamp.
The idea behind measuring each inter¬
val is that the decision where to move
is made before the examinee's hand is
lifted. Therefore, decision time reflects
the time required by the cognitive pro¬
cess of selecting a response. Movement
time reflects a motor movement. Accord¬
ingly, decision time should be a bet¬
ter measurement of cognitive processing
speed than movement time. The analysis
assumes that the independent serial pro¬
cessing assumptions are true.
When people are asked to choose
between k alternatives using an appara¬
tus like Jensen’s, after an initial practice
period , their reaction time is a linear func¬
tion of the logarithm of the number of
choices,
RT[k) = A+B ln(k)t
where A and B are positive real num¬
bers. This is an interesting observation
in itself, because it is the function that
would describe the operation of a deci¬
sion maker who was choosing between
alternatives by making an optimally effi¬
cient sequence of binary choices between
subsets of the alternatives. Therefore, B ,
the slope parameter, can be thought of as
an estimate of the time it takes to decide
which of two equally probable alterna¬
tives has occurred. The A parameter, by
contrast, reflects general speed of pro¬
cessing, including movement time.
Both Sternberg's task and Jensen’s task
are based on Donders’s reasoning. In each
case the argument is that reaction times
increase regularly with increases in the
complexity of the decision process. The
increase in time should be smaller for
more intelligent people.
* Jensen, 2006. This reference summarizes over
twenty-five years of research using the device.
in making a response. Because the reaction
times obtained in identification tasks mix
times for cognitive and motor processes,
the times themselves are not easily inter¬
pretable. Somewhat surprisingly, though,
the way these tasks are used avoids the inter¬
pretation process.
In cognitive psychology, without regard
to individual differences, the typical exper¬
iment focuses on how experimental con¬
ditions change the reaction times in an
identification task. For instance, one of the
major uses of the lexical identification task
is to study how observation of semantically
related items changes the speed of identifi¬
cation of a target item. To take a frequently
cited example, the word NURSE will be rec¬
ognized faster if the immediately preceding
word was DOCTOR than it will if it was
BUTTER.
A similar logic applies to studies of indi¬
vidual differences. When we study the rela¬
tion of elementary cognitive processes to
intelligence, we are interested in the dif¬
ferences in reaction times between people
who have been indexed by their scores on
THE MECHANICS OF INTELLIGENCE
intelligence tests, or by some other measure
of intelligence. The correlation coefficient
is a measure of these differences; if people
who have high intelligence scores are con¬
sistently faster in lexical identification tasks
than people who have low scores, then this
is evidence that at least one of the processes
required for lexical or semantic identifica¬
tion is associated with intelligence. And this
is the case. In the population of college stu¬
dents lexical identification RT has a corre¬
lation of —.40 with scores on tests of ver¬
bal intelligence.11 The negative relation is
expected because short reaction times are
associated with high test scores. The magni¬
tude of the correlation in a more hetero¬
geneous population than college students
would presumably be higher.
Nevertheless, it would be nice to obtain
as direct a measure as possible of cognitive
processing speed. To see this, consider some
facts obtained from general experimental
psychology. In the spirit of Donders, we can
think of any cognitive process as consisting
of the following steps:
Perceive the stimulus -> Select a response
— > Make a response.
The middle step is what we normally
think of as cognition; the first step reflects
the sensory-perceptual apparatus, and the
third step reflects motor processes.
In a modern information-processing
study the participant reacts to information
presented on a computer screen by pressing
one of the keys on the keyboard. Studies in
perception suggest that it takes from twenty
to fifty milliseconds or less to detect a sim¬
ple visual figure, depending on the illumina¬
tion and the complexity of the surrounding
visual field. It takes a healthy college stu¬
dent about 250 milliseconds to press a but¬
ton indicating detection of a visual signal, in
a situation in which the same button is to be
pressed regardless of the identity of the stim¬
ulus. This condition minimizes the middle,
cognitive step in the sequence. If a cognitive
task is introduced, such as asking the student
11 Palmer et al., 1985.
*5*
to recognize whether or not CAMEL is a
word, reaction time rises to 500-600 mil¬
liseconds. This means that of the roughly
550 milliseconds taken for the decision task
somewhere between 250-300 milliseconds,
about half the response time, will be taken
up by perceptual and motor processes. The
other half is taken up by cognitive processes.
Therefore, the reaction time provides a mea¬
sure of cognitive processing speed, albeit a
contaminated one.
This analysis ignores an important issue.
In general, the longer a person waits to make
a response in a choice task, the more likely
they are to make the correct choice. This is
called the speed- accuracy trade-off. Different
people adopt different criteria for accuracy;
one person may be satisfied if his or her
responses are correct on 95% of the trials,
others may prefer 98%, a third may accept
92%. Studies of the speed-accuracy tradeoff
have shown that above accuracy rates of 90%
small differences in accuracy may be asso¬
ciated with substantial differences in reac¬
tion time. Therefore, if different participants
adopt different criteria for acceptable accu¬
racy another source of variation has been
introduced, further confusing the interpre¬
tation of correlations between CRT and IQ
scores.
6.2.3. The Inspection T ime Paradigm
Most of the problems that arise in inter¬
preting CRTs arise because a reaction time
study lumps cognitive and motor process¬
ing together. These problems are avoided
in the inspection time task. In the typical
visual inspection time task two vertical lines
are presented for a brief time, followed by
a visual mask. The interval between the
onset of the two lines and the onset of the
mask is called the inspection time. Figure 6.2
illustrates the procedure. The participant
then indicates whether the line on the right
was longer than the one on the left or vice
versa. The experimenter adjusts the inspec¬
tion time until it is as short as it can be,
while allowing the observer to maintain a
fixed level of accuracy. This is usually set at
75% correct responding.
152
HUMAN INTELLIGENCE
n n
Time 1 1 *
Time A Time B
Figure 6.2. The inspection time paradigm. Two
lines of unequal length are presented at time A,
and displayed until time B . At time B the lines
are replaced by a visual mask blanking out the
stimulus. The observer then indicates which
vertical line was longer. The time interval A-B is
the inspection time The experimenter adjusts
the inspection time until the shortest time
interval is found at which the observer can
maintain a fixed level of accuracy, usually 75 to
80% correct.
Inspection times vary substantially,
depending upon the apparatus, the transient
illumination, and the age and intelligence
of the participants. Some young adults can
maintain accuracy at an inspection time of
30 milliseconds. Senior citizens may pro¬
duce inspection times of 80 to 90 millisec¬
onds. Within age groups and experimental
conditions inspection times are negatively
correlated with intelligence, that is, higher
test scores are associated with lower reac¬
tion times. Correlations in the —.3 to —.4
range are typical. Various statistical correc¬
tions may raise this figure to —.6 to —.7
The inspection time paradigm avoids
mixing cognitive process and motor
responding. However, it does so at the
expense of minimizing the cognitive pro¬
cesses involved, for the task is largely a per¬
ceptual one. Therefore, it is not surprising
to find that when inspection time measures
are included in an analysis of an intelli¬
gence test battery, the inspection time mea¬
sures load on a “cognitive speediness” factor.
This is a broad second-order factor in the
three-stratum model. It has been suggested
that if we broaden the study beyond analy¬
sis of results from young adults, inspection
time might also indicate an ability to control
attention.12
12 Nettlebeck, 2001.
6.2.4. The Relation between Processing
Speed and Intelligence
The correlations between intelligence test
scores and various reaction and inspection
time measures are typically in the —.2 to —.4
range, which is enough to show that there
is a common core to the test scores and
information-processing measures, but small
enough to show that speed of mental pro¬
cessing and psychometric intelligence are
not identical. If the correlations are cor¬
rected for various mediating effects, such as
the restricted range of intelligence test scores
found in university undergraduates, the cor¬
relations may climb to the —.5 to —.7 range,
but the points about overlap and noniden¬
tity remain.
This is reasonable. Early in the mod¬
ern study of individual differences on
information-processing tasks my colleagues
Nancy Frost and Clifford Lunneborg and
I pointed out that the study of individ¬
ual differences in information processing is
quite different from the study of psychome-
trically defined intelligence.13 We saw the
two as related, but not identical, enterprises.
Therefore, we pointed out that the success
of the endeavor would come from combin¬
ing understandings of individual difference
in intelligence, conceptually defined, from the
two different perspectives of cognitive psy¬
chology and psychometric psychology.
The testing procedures required to eval¬
uate elementary cognitive tasks are time-
consuming and expensive. While there are a
few laboratories where banks of computer-
controlled test stations have been built, for
the most part this research takes place using
participant groups that range from a single
individual to perhaps half a dozen people.
Partly for this reason most of the research
on individual differences in elementary cog¬
nitive tasks has been conducted using uni¬
versity undergraduates as participants.
The fact that so many studies are con¬
ducted on populations of young adults
has led to contradictory results and some
13 Hunt, Frost, & Lunneborg, 1973.
THE MECHANICS OF INTELLIGENCE
*53
confusion in the literature. For instance; in a
very well-done study of Spanish undergrad¬
uates; Roberto Colom and his colleagues
found no reliable relationship between
measures of information-processing speed
and intelligence test scores. When they
extended their study population to include
high school students, a reliable relation¬
ship appeared. This study is notable because
exactly the same procedures were used in
each experiment.14 An American experi¬
ment that contrasted information process¬
ing in university undergraduates and high-
level mentally retarded patients found that
both the evidence for a pervasive speed fac¬
tor and the relation of that factor to intel¬
ligence test scores were much stronger in
the mentally challenged group than in the
university undergraduates.15 In perhaps the
largest study done to date, a Scottish repre¬
sentative survey of people from their late
teenage years to their sixties found cor¬
relations between reaction times and per¬
formance on a complex arithmetic task in
the -.3 to -.4 range.16
The potential size of the effects achieved
by expanding the study of the information
processing-intelligence relationship beyond
the ubiquitous college population should
not be underestimated. If we compare “fast”
populations, such as college students, to
“slower” populations, such as the elderly or
mildly mentally retarded individuals, esti¬
mates of processing speed may vary by any¬
where from 50% to a factor of five or more,
depending upon the apparatus and parame¬
ter involved.17
There are also a number of other rea¬
sons for believing that current studies, done
mostly on undergraduates, underestimate
the strength of association between mea¬
sures of speed of information processing and
intelligence test scores. A few of them are
discussed in panel 6.5.
14 Colom et al., 2008. See particularly the contrast
between the results of experiment 1 compared to
experiments 2 and 3.
15 Detterman & Daniel, 1989.
16 Deary & Der, 2005.
17 Ibid.; Hunt, 1980, 1987; Salthouse, 1996.
What are we to make of the moderate
but nontrivial relation between mental pro¬
cessing speed and psychometric measures of
intelligence? Two extreme positions could
be taken. One is that the various elemen¬
tary cognitive tasks evaluate the speeds of
different mental processes, each of which
contributes to cognition. Under this view,
for instance, the lexical identification task
and the inspection time task are measuring
different things. Therefore, each of the pro¬
cesses evaluated should make its own contri¬
bution to the prediction of performance on
complex reasoning tasks, such as are present
on intelligence tests. Another implication of
this view is that any measure of processing
speed that contains a substantial motor com¬
ponent should have markedly lower corre¬
lations with complex reasoning tasks than
would measures associated largely with cog¬
nitive processes.
At the other extreme, it might be that
all of the different information-processing
measures reflect a fundamental speediness
property of a person’s nervous system. If
this is the case, all the different information¬
processing measures would be regarded as
measures of the same pervasive property
and would have similar correlations with test
scores.
The truth seems to lie somewhere in
between. In general, overall measures of
performance, such as the mean CRT in
Jensen’s paradigm, tend to have somewhat
higher correlations with scores on intelli¬
gence tests and complex reasoning tasks
than do derived measures of basic process¬
ing, such as the slope parameter, although
the latter measures should, in theory, eval¬
uate a specific cognitive process. In fact,
the most important variable seems to be
the total time required to complete the
information-processing task. The longer the
time required, the higher the correlation
between performance on the task and on
an intelligence test.18 This is what would be
expected on the assumption that all mea¬
sures tap a single mental speed trait, for then
18 Jensen, 2006.
x54
HUMAN INTELLIGENCE
Panel 6.5. Practical Limits that
Obscure the Relation between
Information-processing Speed
and Intelligence Test Scores
The study of the role of mental¬
processing speed in intelligence is made
difficult by some logistical restrictions.
When cognitive psychologists conduct
reaction time experiments they almost
always provide participants with a sub¬
stantial number of practice trials. This
is done to stabilize performance prior to
measuring such parameters as semantic
access or short-term memory scanning
rates. The reason for doing so is that the
cognitive psychologist wants to be sure
that the procedure is measuring a stable
process in each participant.
Psychologists studying individual dif¬
ferences need to have a large number of
participants in order to obtain adequate
statistical power for the analysis of cor¬
relations. This makes taking great care to
obtain reliable measurements from each
participant an expensive proposition. As
a result, many studies of the relation
between performance on elementary cog¬
nitive tasks and psychometric test scores
are based on studies in which, by the
standards of cognitive psychology, the
participants simply did not have enough
practice.
If the participants are trying to fig¬
ure out how to use the testing appara¬
tus at the time that measurements are
being taken, an investigator could obtain
a spuriously high correlation between
the information-processing measure and
an IQ score, because high-IQ individ¬
uals are quicker at finding out how to
work the apparatus. Alternatively, the
correlation might be spuriously decreased
because individual performance on the
information-processing measure will not
have stabilized at the time of measure¬
ment. All we can say is that this feature
of many experiments makes interpreta¬
tion of the results confusing.
There is also a question of reliability.
Psychometric tests of intelligence have
internal consistency statistics (split-half
reliabilities or Cronbach's a statistic) in
the .80-.95 range. Reaction time (RT)
measures typically have even higher anal¬
ogous correlations, often in the .90-.95
range. Within a test session RT measures
are highly reliable.
Things are quite different if we exam¬
ine reliability across sessions. The cor¬
relation between equivalent measures of
a conventional psychometric intelligence
test, taken a few days apart, will typically
fall in the .8-.g range. Taken together
with the internal consistency statistics,
this shows that the trait being measured
is quite stable over intervals of days or
weeks. RT measures show much greater
day-to-day variability. Test-rest correla¬
tions can drop to .6 at intervals of only
one or two days.* Because the within-
day reliability is much higher, the trait
being measured must itself be variable
from day to day. Such variability con¬
strains the upper value of the correlation
between an RT measure and a psycho-
metrically measured trait. If we correct
for this effect statistically, the estimated
correlation between the traits mea¬
sured by information-processing mea¬
sures and intelligence test scores jumps
from an observed .3-. 4 to an estimated
■6~7*
It would be possible to conduct a study
in which RT measures were taken over
several days, and the trait common to
them was correlated with psychometric
test scores. Such a study would be fairly
expensive and, as far as I know, has not
been done.
* Bittner et al., 1986.
THE MECHANICS OF INTELLIGENCE
*55
the longer tasks would provide a better mea¬
sure of individual differences in processing
speed.
However; different tasks do have some
specificity. For instance, lexical decision
times correlate most highly with verbal
intelligence measures, while inspection time
measures correlate most highly with nonver¬
bal tests. Nevertheless, the general trend is
clear. General cognitive speediness is a reli¬
able trait, and is certainly a part of what we
mean by “intelligence.”
Some critics of research on mental pro¬
cessing speed have attacked the idea that
“it is good to be fast” on the grounds there
are situations in which this is not true.19 In
fact, some societies distrust rapid respon¬
ders, believing that the more intelligent indi¬
vidual is the one who stops to weigh alter¬
natives before speaking.
This objection misses the point of the
research. Studies of individual differences in
processing speed are concerned with how
rapidly a person can grasp a situation, given
a fixed amount of information. Deciding to
make an overt response is a separate action.
Being a bit flippant, there is a difference
between noticing the logical flaw in your
father-in-law’s political views and deciding
to point out that flaw to him. These two
separate acts of cognition both benefit from
rapid thinking. Indeed, rapid thinking will
probably help you suppress inappropriate
or impolitic responses in many social set¬
tings. The assertion that there is an associa¬
tion between speed of information process¬
ing, measured by simple cognitive tasks, and
the ability to reason in complex settings is
not equivalent to saying that the intelligent
individual necessarily has the fastest mouth
in town.
6.3. Working Memory and General
Intelligence
One of my colleagues once observed that
since time is continuous the present just
19 See, e.g., Sternberg, 1996.
became the past. Therefore, perception is
impossible; all psychology reduces to mem¬
ory and imagination.20 Leaving aside imag¬
ination, a great deal of our thinking does
depend on memory.
Psychologists distinguish between two
types of memory: immediate memory,
which spans at most a few seconds, and
long-term memory, in principle spanning
the individual’s lifetime, although retrieval
becomes difficult as time passes. The Black¬
board Model uses this distinction strongly.
The blackboard is itself an immediate mem¬
ory device, and is thought of as being a
severely limited resource, while long-term
memory is conceived of as being essentially
limitless. We now want to take a closer look
at the structure of the blackboard.
Battery- type intelligence tests, such as the
WAIS, often contain separate subtests for
short-term memory and long-term memory.
Short-term memory is commonly evaluated
by a digit span test. In a digit span test the
examiner presents a randomly chosen series
of three to eight digits, for example, “3, 9,
7, 4.” The examinee’s task is to echo the
digits immediately after they have been pre¬
sented. In the somewhat more challenging
backward digit span task, the examinee must
echo the digits in reverse order, “4, 7, 9, 3”
in the example.
The forward digit span task is generally
regarded as a test of a person’s short-term
storage capacity. Digit span tests typically
have a loading of about .5 on the general
factor of an intelligence test battery.21 Back¬
ward digit span introduces a processing as
well as a memory component; instead of
“reading out” memory the examinee has to
reverse the order in which information has
been received. The backward digit span is
more highly correlated with total test scores
than the forward span. This suggests that we
ought not regard the blackboard-immediate
memory system as just a storage system. Nor
shall we.
20 Poltrock, 1977.
21 See Gignac, 2006, and O’ Grady, 1983, for results on
the WAIS, spanning more than twenty years.
i56
HUMAN INTELLIGENCE
When Alan Baddeley introduced the idea
of working memory,22 he pointed out that
when people solve problems they have to
hold key pieces of information readily at
hand, even though they are not at the imme¬
diate focus of consciousness. In addition,
information has to be transformed, men¬
tal representations have to be built, and
attention has to be focused on informa¬
tion relevant to the task at hand, while
irrelevant information must be suppressed.
Baddeley’s view of working memory, which
is incorporated into the Blackboard Model,
is that it is a mental workspace containing (a)
modality-specific, passive memories [slave
memories , in Baddeley’s terminology) for
small amounts of auditory and spatial-visual
information and (b) a somewhat underspec¬
ified “executive” that supervises the flow in
and out of the storage systems. Variants of
this model that include an additional slave
memory for abstract semantic information
have since been proposed.25
Baddeley’s “central executive” was not
well specified. At the least, there has to be
provision for a process that focuses attention
on some information in working memory, to
make it available for processing, while at the
same time preventing [total) loss of infor¬
mation that is not currently at the focus of
attention, but that has been accessed in the
recent past and may be needed in the imme¬
diate future.
This idea is encompassed by the term
“situation awareness,” a phrase that has
become popular in engineering psychology.
The term refers to the need to be aware of
what is going on about you, even if some
aspects of the situation are not of imme¬
diate concern. For instance, an automobile
driver has to keep aware of the positions
of nearby cars, including those behind the
driver's vehicle and hence not in immediate
view.
Similar requirements occur in many sit¬
uations, some quite far removed from driv¬
ing. When attorneys question witnesses dur¬
ing criminal trials they have to be aware
22 Baddeley, 1986; Baddeley & Hitch, 1974.
23 Hunt, 2002.
of jurors’ reactions as they are interrogat¬
ing the witness. When people attack prob¬
lems in logic they have to be aware of differ¬
ent interpretations of the premises. When
chess players consider a move they must
evaluate different counters that the oppo¬
nent might make. From an information¬
processing view, controlled manipulation
of information in working memory is an
extremely important part of cognition.
Therefore, we would expect individual dif¬
ferences in the effectiveness of the work¬
ing memory system to be related to general
intelligence. And they are.
6.3.1. Measuring the Capacity of
Working Memory
The span task is a popular technique for
evaluating working memory capacity. In a
span task a person is asked to perform a
series of simple actions [e.g., reading a sen¬
tence) while holding more and more infor¬
mation in memory as each sentence is read.
The simple task will be called the primary
task; the act of holding information in mem¬
ory is the secondary task. The idea is that
the processing requirements of the primary
task interfere with holding the information
needed for the secondary task. Therefore the
amount of information that can be held in
memory in the face of the periodic inter¬
ruptions by the primary task provides a
measure of the storage capacity of working
memory.
Two typical span tasks are shown in
panel 6.6. When the primary task is reading,
the span measure has a .5 correlation with
measures of verbal comprehension, includ¬
ing the paragraph comprehension tasks
that are often found on tests of verbal
intelligence.24
When different types of span tasks are
included in a comprehensive battery of rea¬
soning and information-processing tasks it
turns out that (a) all the span tasks reflect
a common underlying factor, referred to
as working memory capacity (WMC), and
[b) WMC has a correlation of .64 with a
24 Daneman & Merikle, 1996.
THE MECHANICS OF INTELLIGENCE
*57
Panel 6.6. Memory Span Tasks
The first, and still the most widely used,
memory span task is the reading span task
developed by two Camegie-Mellon Uni¬
versity researchers, Meredyth Daneman
and Patricia Carpenter.* The examinee
hears or is shown from one to five unre¬
lated sentences, one at a time, and then
is asked to repeat the last word of each
sentence. An example is
He walked down Michigan Avenue in
the face of the bitter wind.
The restaurant was known for its cre¬
ative preparation offish dishes.
All the cabinet wished the foreign secre¬
tary success on her vital trip.
RECALL
The examinee is then supposed to
recall wind, dishes, trip.
Randolph Engle, a Georgia Institute of
Technology professor, and his colleagues
have generalized this finding.7 They point
out that, abstractly, the span task requires
processing some information (the sen¬
tences) while holding other information
(the words) in abeyance. Therefore, a
(somewhat) nonverbal span task, process¬
ing span, can be created. In the following
example the task is to read and verify the
arithmetic equation, and then store the
word that follows the answer.
The examinee is shown
IS (8/4) -1=1?
The examinee responds “yes.” The
word bear is then presented. This is fol¬
lowed by the following sequences:
IS (6*2) — 2 = 10? (Response: YES)
Beans
IS (10*2) -6 =s 12? (Response: NO)
Dad
RECALL
The examinee replies “bear, beans,
dad A
Span tasks similar in structure can be
constructed for a variety of operations in
other domains, such as counting, keeping
track of movements, and imagined rota¬
tion of visual objects.
* Daneman & Carpenter, 1980.
t Engle etal., 1999; Kane & Engle, 2002; Kane etal.,
2001, 2004.
general reasoning factor extracted from con¬
ventional intelligence tests.25
Span tasks are a special case of dual tasks,
tasks in which people are asked to attend to
at least two streams of information at once.
A person who listens to the radio while driv¬
ing a car is performing a dual task. Many lab¬
oratory tasks are formalizations of this every¬
day paradigm. In a dual task attention has to
be switched from one stream of information
to another, and then back again, with mini¬
mum loss of information during the switch.
Dual tasks call on both the storage and atten-
tional control aspects of working memory.
In a widely referenced series of studies by
Patrick Kyllonen and his colleagues at the
US Air Force’s Armstrong Laboratory, Air
25 Kane et al., 2001.
Force recruits performed a number of work¬
ing memory tasks, including span tasks, and
also took highly g-loaded tests of reasoning.
The recruits’ scores on the ASVAB were
also available. A common factor extracted
from the working memory tasks predicted
virtually all the variance extracted from the
reasoning tests and from the general factor
on the ASVAB test.26 Similar findings have
been reported by other investigators. This
work is important for two reasons. First,
the populations studied included US Air
Force enlistees, thus broadening the studies
beyond the usual range of college students.
(While Air Force enlistees generally do not
have the test scores of college students,
26 Kyllonen & Christal, 1990; Kyllonen & Stephens,
199°.
i58
HUMAN INTELLIGENCE
they are of somewhat above-average ability
compared to the general population.) Sec¬
ond, the number of different working mem¬
ory tasks used ensured that the result did
not depend upon the unique information¬
processing requirements of particular tasks.
The latter finding poses something of an
intellectual challenge. As the Spanish psy¬
chologist Roberto Colom put it;
Working memory comprises the functions
of focusing attention, conscious rehearsal,
and transformation and mental manipu¬
lation of information.
Colom et al, 2004, p. 277
Kyllonen’s work and the studies that fol¬
lowed it show that working memory a
complex information-processing concept, is
related to general intelligence. But what part
of working memory is crucial? Ian Deary, a
professor at the University of Edinburgh,27
has pointed out that we do not advance
understanding by showing that one mysteri¬
ous concept is linked to another.
There are two ways to reply to Deary’s
objection. One is to show that one, two, or
all of the information-processing functions
that comprise working memory make sepa¬
rate contributions to the relationship with g.
That is, we reduce the intelligence-working
memory link to links between intelligence
and the component parts of working mem¬
ory. The other is to accept the linkage at
the intelligence-working memory level, and
challenge the need for further reduction.
In order to discriminate between these
possibilities we need to have separate mea¬
sures of Colom and colleagues’ four aspects
of working memory. We require:
(a) Memory span tests, to evaluate the abil¬
ity to store information while processing
another concurrent task.
(b) Tests that evaluate short-term storage
without processing - for instance, recall
of a short string of digits without any
intervening or concurrent processing,
(c) Tests that evaluate the ability to con¬
trol attention without any memory
27 Deary, 2000.
component. An example of such a test
is shown in panel 6.7.
(d) Tests that evaluate processing speed,
such as a choice reaction time task. This
is considered necessary because speed
of stimulus identification, decision mak¬
ing, and rehearsal of information are
all involved in the working memory
tasks.
(e) One or more tests shown to be markers
of reasoning, fluid intelligence, or gen¬
eral intelligence.
Structural equation modeling can then be
used to identify latent traits underlying each
of the five types of tests, and to determine
the relationships between the traits.
Several such studies have been done, with
somewhat confusing results. Within each
of the studies a coherent picture emerges,
often identifying a key component deter¬
mining the relation between working mem¬
ory and intelligence. Across studies there
is very little agreement over what the
component is.
The array of studies has an international
flavor. Studies of American university stu¬
dents produce results indicating that the
key component is the ability to control
attention.28 Studies of Spanish high school
and university students29 indicate that the
storage process is the key component. Stud¬
ies with German university students pro¬
duce results pointing toward “supervision,”
which is described as the ability to keep
track of the current state of variables that
change over time - for example, the location
of the car in the driving example given ear¬
lier. (This process is sometimes referred to
as updating .)?°
It seems unlikely that working memory
would work differently in Spain, Germany,
and the United States. And just in case you
think it might, a study that explicitly com¬
pared the architecture of information pro¬
cessing in two different countries showed
28 Engle et al., 1999; Heitz, Unsworth, & Engle, 2005;
Kane et al., 2001, 2004.
29 Colom et al., 2008.
30 Oberauer et al., 2003.
THE MECHANICS OF INTELLIGENCE
159
Panel 6.7. An Attention-switching
Task
Several different tasks have been devised
to measure the ability to switch atten¬
tion from one concurrent task to
another. They all follow this general
outline.
1. The observer is presented with two
simultaneously presented streams of
stimuli, and is told to search for tar¬
gets on one. Stimuli are presented
rapidly.
2. Aperiodically the observer receives a
signal indicating that he or she should
either continue to search for targets
on the stream being monitored or
shift to the other stream.
3. The dependent variable is the num¬
ber of targets missed immediately
after the signal is sounded. This
is taken as a measure of the time
required to fix attention on the new
target stream.
The following example illustrates the
procedure. The observer is told that he
or she will see a sequence of pairs of
letters presented on a computer screen.
The letters will be presented reason¬
ably far apart, but both letters will be
within the visual field. Only one of these
sequences is to be monitored at any one
time.
The task is to press a button whenever
a vowel appears in the position (right or
left) currently being monitored.
At the start, monitor the right-hand
sequence.
Subsequently, if a high tone is
sounded, switch from the currently mon¬
itored sequence to the other (e.g., from
the right to the left). If a low tone is
sounded, continue to monitor the se¬
quence being monitored before the tone
was sounded.
Here is an example. Each line repre¬
sents a time period.
Left-hand Right-hand
Sequence Sequence
R
L
K
X
O
R
Z
E (target)
T
K
(High tone sounds)
(target)
I
J
N
S
(Low tone sounds)
Q
Y
(target)
u
C
Doing well on this task hardly qualifies
as intelligence, in the conventional mean¬
ing of the word. However, the ability to
switch back and forth from one stream
of information to another can be vital
in certain jobs. Air traffic controller is a
good example. Empirically, the ability to
perform well on attention-switching tasks
is statistically associated with scores on
measures of fluid intelligence.*
* Duncan et a!., 1996.
that the overall relations between structures
were the same in each. The countries com¬
pared were Greece and China!31
In spite of the confusing results, I
believe the findings have a rather simple
31 Demetriou et al., 2005. Chinese children and ado¬
lescents were better than Greeks at tasks involving
visual-spatial reasoning, but there was no difference
in variables specifying the relation between working
memory and reasoning.
explanation. There is little doubt that work¬
ing memory is an important component of
general intelligence, although the two are
not identical, either in the sense of being
perfectly correlated with each other statis¬
tically or in the sense of being conceptu¬
ally identical.32 It is not surprising to find
32 In addition to the studies cited here, my conclusion
is based on an excellent review of the literature by
Ackerman, Beier, & Boyle (2005}.
i6o
HUMAN INTELLIGENCE
that the constraining features of working
memory may differ in different popula¬
tions. For instance, studies in the ubiqui¬
tous university student population show rel¬
atively small relations between processing
speed and intelligence test performance. If
we extend the studies over the full adult
range processing speed becomes an impor¬
tant variable. Finally, confusion may result
because studies that try to isolate the most
important aspect of the working memory-
intelligence relationship may be trying to
carry the analysis to a finer level of detail
than is appropriate. I now elaborate on this
point.
6.3.2. Brunsivickian Symmetry:
Working Memory and Intelligence
as System Effects
In a talk given at the University of Wash¬
ington, the German psychologist Werner
Wittmann made the point that when we
compare two sets of behaviors it is important
to maintain the same level of complexity for
each behavior. Wittman referred to the bal¬
ance as Brunsivickian Symmetry , by analogy
to some of the ideas of Egon Brunswick con¬
cerning decision making. Wittmann’s argu¬
ment can profitably be applied to studies of
the relation between working memory and
general intelligence.
The terms general intelligence and work¬
ing memory both refer to systems of psy¬
chological functions, rather than individual
functions. Consider the case of what many
consider to be the best single measure of
general intelligence, the progressive matrix
item. A moderately difficult item of this sort
is shown in Figure 6.3. What does it take to
solve this sort of problem?
You have to be able to pick apart the
individual elements of the figures, recogniz¬
ing that each figure has certain attributes,
rather than reacting to the figure as a whole.
You have to be able to recognize that
each attribute of a component of the fig¬
ure appears once in each row and that each
attribute of another component appears
once in each column. Because visual atten¬
tion is limited, you must be able to hold
in memory information about recognizing
an item in a row as you look down the
rows and across the columns. Then you have
to fill in the missing pattern. Finally, you
have to hold the correct pattern in short¬
term memory as you search the alternative
answers provided, until you find one that
matches the pattern. There is no one neces¬
sary working memory function. All of them
are required.
Working memory and intelligence are
both concepts that refer to the ability to
coordinate a system of elementary functions
into a coherent whole. There are relations
at the system level, but trying to establish
relations between the elementary functions,
below the level of overall system function¬
ing, is not likely to be useful. The quest for a
single information-processing function that
explains intelligence is no more likely to suc¬
ceed than the quest for the Holy Grail.35
6.4. Verbal Comprehension
Verbal comprehension is the process of under¬
standing written or spoken communication.
In adult readers, the two processes are highly
correlated.34 Comprehension is somewhat
different from production, although provid¬
ing that one does not demand elegance in
writing or speaking, the abilities to under¬
stand and to generate linguistic messages
are closely related. Tests of verbal compre¬
hension are contained in almost all battery-
type intelligence tests. For instance, people
who take the SAT have to show that they
can extract meaning from passages that are
roughly the length of a newspaper editorial,
and that deal with serious topics.
Over the general range of ability, under¬
standing language and general intelligence
are closely intertwined, although not per¬
fectly correlated. In extreme cases the pro¬
cesses do become distinct; poets are not
necessarily good at high level mathemati¬
cal reasoning, nor are mathematicians nec¬
essarily literary virtuosos. However, it is
33 Hunt, 1987.
34 Palmer et al., 1985.
THE MECHANICS OF INTELLIGENCE
161
☆
1=3
☆ ☆
a
☆
tr
☆ ☆
c=0
☆
a
☆ ☆
tr
☆ ☆ ☆
tr
☆☆ ☆
a
☆
|=3
☆ ☆
tr
☆☆ ☆
☆ ☆
☆ ☆ ☆
☆☆ ☆
☆
☆☆ ☆
tr
tr
tr
a
Figure 6.3. A progressive matrix item of intermediate difficulty.
The task is to decide which of the figures below the line is needed
to replace the blank item in the 3x3 matrix above the line. The
problem was constructed as a demonstration, using Carpenter, Just,
and ShelFs (1990) rules for constructing progressive matrix items.
important to keep this distinction in per¬
spective. People who have an elegant com¬
mand of the language do not display
incompetence in logical and mathematical
reasoning, nor are logicians and mathemati¬
cians likely to be inarticulate.
At the other end of the intelligence scale,
low general intelligence is almost always
associated with a pervasive low level of ver¬
bal comprehension. On occasion, though,
we find a disassociation between general rea¬
soning and some aspects of verbal skill. Here
is an interesting example.
Williams syndrome is a genetically deter¬
mined type of mild to moderate men¬
tal retardation. Patients with Williams syn¬
drome display language abilities that are
considerably higher than their general rea¬
soning abilities. At one time it was thought
that this was evidence for a disassociation
between linguistic skills and general cogni¬
tion, but further research has shown that
that is not quite accurate. Williams syn¬
drome patients show surprising verbal skill,
compared to other people suffering from
the same level of general mental deficiency.
However, their use of language lags behind
that of age-matched children with normal
mental capacities. One of the most interest¬
ing deficiencies is their tendency to react to
the literal meaning of statements, thus fail¬
ing to grasp the meaning of metaphorical
statements or irony.35 A Williams syndrome
patient might profoundly misinterpret state¬
ments like He’s full of baloney or I suppose
he’ll relax by going skydiving.
To go further we have to consider
what language comprehension is. Walter
Kintsch, a University of Colorado profes¬
sor, has developed a useful framework for
understanding the comprehension process.36
Kintsch distinguished three stages in com¬
prehension. First there are the purely lin¬
guistic processes of retrieving word mean¬
ing ( lexical retrieval, in technical terms)
and applying syntactic rules in order to
extract meaning from phrases and sentences.
These tasks comprise low-level comprehen¬
sion. Low-level comprehension is followed
by high-level comprehension, in which sen¬
tence and phrase meanings are incorporated
into a text model, essentially an understand¬
ing of what a text says. As the text model is
built, some information is dropped out and
other information is highlighted. The text
model is then incorporated into a situation
model, which represents the meaning of the
35 Mervis, 2003.
36 Kintsch, 1998.
162
HUMAN INTELLIGENCE
text, in the light of other information that
the comprehender has and sees relevant.
Kintsch presents text comprehension as
a highly interactive process. Sentence com¬
prehension, building the text model, and
building the situation model do not take
place in series; there are many feedback
loops. Going into the theory in detail would
be rather far afield from our concern with
individual differences. Instead I offer exam¬
ples in the spirit of Kintsch ’s analysis, with¬
out going into details.
The following statement was made by
Senator John McCain, about Senator Hillary
Clinton, at a time when McCain was a can¬
didate for the 2008 Republican presidential
nomination and Clinton was a candidate for
the Democratic nomination.
In case you missed it, a jew days ago,
Senator Clinton tried to spend $1 million
on the Woodstock Concert Museum. Now,
my friends, I wasn't there. Ym sure it was
a cultural and pharmaceutical event ... 7
was tied up at the time}1
Senator John McCain, Oct. 29,
2007, at a Republican debate in
Orlando, Florida
The following propositions make up my
(informal) text base
Senator Clinton tried to spend a million
dollars.
The money was for the Woodstock Concert
Museum.
McCain was not at ( the Woodstock
concert).
McCain believes the concert was a cultural
and pharmaceutical event.
McCain was occupied with other matters
at the time.
Certain statements in the text are unim¬
portant - for example, the qualifier “in
case you missed it.” They drop out of the
text base. Strictly speaking, the statement
“I wasn’t there” is ambiguous; it could refer
37 McCain’s statement was accurate. Clinton had
joined with New York’s other senator, Charles
Schumer, in an attempt to secure funds for the
museum.
either to the concert or to the time when
Senator Clinton tried to acquire money for
the museum.
The situation model puts this in the con¬
text of unstated but widely known infor¬
mation about the Woodstock concert and
about McCain himself. Of course, such a
model depends upon the comprehender s
knowledge. Here is what my situation model
is. Information not specifically stated in the
text base is shown in parentheses.
Clinton wants to appropriate one million
dollars for a museum commemorating the
Woodstock concert. (Government money,
not her own.)
The Woodstock concert was a cultural
event. (It was the icon of the radical youth
culture of the late 19 60s and early 1970s.)
There was a great deal of (illegal) drug use
at the concert.
At the time of the concert McCain was not
present. (He was a prisoner of war in Viet¬
nam, where he was badly treated, i.e., he
was literally tied up.)
An even wider situation model assumes
that McCain was contrasting his undeni¬
ably patriotic record during the Vietnam
War to Clinton’s attempt to memorialize
an event that many in his Republican audi¬
ence considered an example of moral decay
during the “hippy” era. Other readers might
develop other models. In each case the situa¬
tion model will depend heavily upon knowl¬
edge that the comprehender brings to bear.
This depends upon both what the com¬
prehender knows and what he or she sees
as relevant to understanding the current
statement.
Our interest is in the role of individual
differences in information processing during
both the low-level stage of understanding
what the text literally says and the high-level
stage of understanding what it means.
6.4.1. Information Processing and
Low-Level Linguistic Skills
Every normal human being acquires a com¬
plex set of rules for speaking and listening,
THE MECHANICS OF INTELLIGENCE
l63
up to the level of analysis of sentences and
phrases, in a stunning display of tacit learn¬
ing. But we are not all equally adept lin¬
guists. There are subtle differences among
individuals in their ability to understand
language.
Different people know different words.
A score on a brief vocabulary test is a
guide to a person's general intelligence. 38
For instance, the vocabulary test included
in the WAIS has a loading of .80 on the
general factor.39 Having a small vocabulary
does not necessarily reflect a deficiency in
information-processing capability, for dif¬
ferences in vocabulary may simply reflect
differences in a person's social environment.
Word knowledge, in itself, is part of intel¬
ligence, for your vocabulary does, in part,
determine how well you can function in
society. Just knowing words, however, is
not an information-processing component
of intelligence.
People differ in the speed with which
they can retrieve well-known words, which
is an information-processing capability. Peo¬
ple who have high scores on tests of verbal
comprehension recognize common words
more rapidly than people with low test
scores.40 They also recognize well-known
semantic relationships, answering questions
like “Is a deer an animal?” more rapidly
than their low-scoring counterparts.41 This
appears to be partly a general processing-
speed effect and partly a more restricted ver¬
bal skill.
There is one piece of the mechanics of
word recognition that is not related to verbal
comprehension test scores. In general, peo¬
ple are quicker to recognize a word when
it is presented shortly after a related word.
The word NURSE will be recognized more
quickly if a person is first asked to recognize
a related word, like DOCTOR, than if an
unrelated word, like BUTTER, is presented
first. This phenomenon is called semantic
priming. While priming is obviously useful in
verbal comprehension, there are only small
38 Carroll, 1993.
39 Gignac, 2006.
40 Palmer et al.( 1985.
41 Goldberg, Schwartz, & Stewart, 1977.
individual differences in priming, and they
are not related to individual differences in
verbal comprehension.42
Moving from word recognition to sen¬
tence recognition, we find both effects of
accuracy of lexical retrieval and effects of
working memory. Marcel Just and Patri¬
cia Carpenter43 measured people’s work¬
ing memory, using the reading span task
described earlier. They then determined the
time that their participants took to analyze
complex sentences and how accurately they
were able to analyze them. Here are two
illustrations of their work.
Consider the two sentences
The evidence examined by the lawyer
shocked the jury.
The defendant examined by the lawyer
shocked the jury.
In the first sentence examined must intro¬
duce a relative clause modifying evidence,
because evidence, being inanimate, cannot
examine anything. In the second case exam¬
ined might refer to an activity by the defen¬
dant or it might introduce a relative clause.
The ambiguity cannot be resolved until the
phrase by the lawyer is encountered. At
an absolute level, people with high work¬
ing memory capacity read both sentences
faster than people with low working mem¬
ory capacity. However, people with high
capacity read the first sentence more rapidly
than the second, that is, they took advantage
of the ambiguity resolution afforded by the
inanimate quality of evidence. People with
lower memory spans read both sentences at
about the same speed.
Now consider the following two sen¬
tences:
The experienced soldiers warned about the
raid before the midnight attack.
The experienced soldiers warned about the
raid conducted the midnight attack.
The phrase The experienced soldiers
warned about the raid is ambiguous; it could
42 Palmer et al., 1985.
43 Just & Carpenter, 1992; MacDonald, Just, &
Carpenter, 1992.
164
HUMAN INTELLIGENCE
refer to a warning either given by or given to
the soldiers. The ambiguity is resolved when
the word before or conducted is read. Read¬
ers with high working memory capacities
slowed down when they encountered the
ambiguity-resolving word. People with low
working memory capacities did not. When
asked questions about the interpretation of
the second question, such as Did anyone
warn the soldiers?, people with high spans
were more likely to answer the question
correctly.
Just and Carpenter interpreted these
results as showing that the high-span
(and, by inference, high-working-memory-
capacity] readers carried forward both
meanings of the ambiguous phrase until
the ambiguity was resolved, while the low-
span readers carried forward only the pre¬
ferred alternative, that the soldiers did the
warning.
Just and Carpenter also considered the
effects of an extrinsic task on sentence
comprehension. Recall that in the mem¬
ory span task a person reads a sentence,
stores the last word, then reads another sen¬
tence, and so forth. Memory span is deter¬
mined by the number of words that can
be recalled as more and more sentences are
read. Just and Carpenter turned this proce¬
dure around. First they established people's
memory spans, using a fixed set of sentences.
In a separate procedure, they presented
material to be retained in memory and then
presented sentences of varying complexity,
and determined whether or not the readers
understood them. This gave them a mea¬
sure of sentence-processing capability in the
presence of a concurrent short-term mem¬
ory load. Understanding decreased as sen¬
tence complexity increased, but people with
high spans were less bothered by the concur¬
rent memory load than were people with
low spans.
Just and Carpenter concluded that work¬
ing memory should be considered a capac¬
ity for processing, rather than a reflec¬
tion of either accurate lexical retrieval or
short-term storage capacity. I would put
this somewhat more neutrally. Working
memory is a system of storage, retrieval,
processing, and attention control functions.
The important thing is how well these func¬
tions work together to create a system for
managing information. When dealing with
verbal comprehension we regard working
memory as a component of verbal ability;
when we deal with reasoning more gener¬
ally, we talk about the effect of working
memory on g. In either case individual dif¬
ferences in the effectiveness of the working
memory system are central to individual dif¬
ferences in cognition.
6.4.2. Higher-Order Comprehension
Processes
It is difficult to distinguish higher-order ver¬
bal comprehension from intelligence, for
some of the most complicated items on cog¬
nitive tests are verbal comprehension items.
Doing well on the SAT requires comprehen¬
sion of quite complex statements. It is some¬
what more profitable to present an analysis
of how texts are comprehended, continu¬
ing with an informal use of Kintsch's model.
The analysis will indicate the points at which
information-processing constraints become
important.
The following rather convoluted passage
appeared in the Yale Alumni Magazine .
Whatever fails to accord with the values
of political liberalism fits uncomfortably
within the range of possibilities that the pre¬
vailing conception of diversity permits stu¬
dents to acknowledge as serious contenders
in the search for an answer to the first-
person question of what living is for .
Kronman, 2007, p. 26.**
My first reaction when I read this was
Eh?
All-purpose word in Canadian
English
But, with an effort, this forty-five-word sen¬
tence makes sense. You just have to take it
stage by stage.
Whatever fails to accord with political lib¬
eralism -> nonliberal ideas
44 Kronman is a former Dean of the Yale Law School.
THE MECHANICS OF INTELLIGENCE
165
Parse phrase, refer to long-term memory
for semantic references, and store in
memory.
fits uncomfortably within the range -> is
outside of, not permitted by
Parse phrase, refer to long-term memory
for semantic references, store result in
memory. Working memory now holds
nonliberal ideas not permitted by.
of possibilities that the prevailing conception
of diversity permits -> ideas permitted by
the current concept of diversity
Parse phrase, refer to long-term memory,
identify possibilities with ideas.
Working memory now holds nonliberal
ideas not permitted by the current con¬
cept of diversity.
Next we encounter students. This changes
the presumed grammatical structure of the
sentence. It turns out that “concept of diver¬
sity” is the subject of the sentence, not the
object of the preposition by. We have an
example of updating, one of the functions
of working memory. Appropriate rearrange¬
ment of the contents of working memory
produces
The current concept of diversity does not per¬
mit students
and non-liberal ideas is free floating until
the next word occurs.
to acknowledge
This anchors the term nonliberal ideas.
Working memory now holds
The current concept of diversity does not per¬
mit students to acknowledge nonliberal
ideas.
We next encounter a mouthful (memo¬
ryful?) of relative clauses.
as serious contenders in the search for an
answer to the first-personal question of
what living is for
These require similar parsing and analy¬
sis. I will not go through the steps. The
result is fairly simple.
Current concepts of diversity do not let stu¬
dents consider nonliberal views.
We have gone from forty-five words in
the statement to eleven in the text base. We
can move to the situational model by recall¬
ing that Kronman was writing in 2007, and
that he was addressing Yale alumni. A com¬
pleted part of the situation model is
There has been concern over academic
biases against conservative thought.
Professor Kronman is trying to explain the
current situation to the Yale alumni.
Kronman says that
Views in favor of cultural diversity pre¬
vail on campus.
These views do not permit students to
consider competing ideas.
Perhaps Kronman could have written
more simply. My purpose is not to edit
his writing. It is to show that comprehen¬
sion of language, a very important part
of human intelligence by any definition,
requires a complex interplay between the
working memory and long-term memory
systems. Long-term memory does not just
act as a provider of information and syntac¬
tic rules. It also holds information needed to
understand what the text means in the con¬
text in which it is stated. Neither McCain’s
nor Kronman’ s statement can be understood
unless the comprehender knows, and sees as
relevant, a great deal of cultural information
about American political and social issues in
the early twenty-first century.
The ability to bring such information to
bear is an important part of intelligence. It
has to be evaluated if one wants to make a
serious claim that one’s test measures ver¬
bal intelligence in any meaningful way. This
poses something of a problem to the test
maker. There may be situations in which the
test maker does not want to evaluate pos¬
session of knowledge, because possession of
knowledge depends upon a particular cul¬
tural background. At the same time the test
maker does want to evaluate the ability to
use knowledge to build a situation model.
What to do?
The answer is to be clever. Here is an
elegant example of evaluating an elementary
school student’s ability to build a situation
model. It was a test item in an Australian
i66
HUMAN INTELLIGENCE
examination intended to assess a student’s
ability to understand irony.45
Text: Lovely mosquito, sitting on my arm
Stay right where you are , I mean you no
harm.
Still as a statue, stand right where you're
at.
I only mean to give you a pat.
Question: Does the writer like the
mosquito ?
The best of our computer programs for
automatic language comprehension would
have a hard time answering this question. I
asked a reasonably intelligent fourth grader,
who replied (with an evil grin)
He wants to kill it.
6.5. Information-processing
Mechanisms Underlying Visual-spatial
Reasoning
Visual-spatial reasoning is included in psy¬
chometric models as the Gv (visual intel¬
ligence) second-order factor in the three-
stratum model, and by the perceptual and
rotational dimensions in the g-VPR model.46
The “visual” term is justified because the
ability refers to the ability to detect, recog¬
nize, and analyze objects in the visual field,
and in visual imagery, the ability to manip¬
ulate objects in the mind’s eye. The two
are closely related, although not perfectly
correlated 47 The “spatial” term refers to
the ability to reason about real or imagined
movement in space, including one’s own
position and movement relative to other
objects (orientation). Orientation ability is
not measured in the major battery- type
intelligence tests, but it is evaluated in some
personnel-selection situations, particularly
in aviation.
45 I encountered this while attending a National
Research Council workshop on assessment of stu¬
dent achievement. I reproduce it from memory, and
may not have the verse absolutely correct.
46 Carroll, 1993; Johnson & Bouchard, 2005a.
47 Burton & Fogarty, 2003; Kosslyn, 1980; Poltrock &
Brown, 1984.
Table 6.1. Correlations between different
visual-spatial abilities
Narrow Factor
CF
CS
SR
VZ
cs
.65
SR
.60
•32
VZ
.58
.65
•77
MS
•52
■43
•43
.46
Source: Data excerpted from Burton & Fogarty,
2003, with permission from Elsevier.
An analogous factor, auditory ability
(Ga), has been identified in research on the
three-stratum model.48 Auditory ability was
left out of the g-VPR model simply because
none of the test batteries analyzed by John¬
son and Bouchard contained any auditory
tests. It is difficult to fit tests of audition
into a paper-and-pencil testing paradigm,
but the recent developments in delivery of
computer-controlled music should make it
easier to explore individual differences in
auditory ability in the future.
A study by two Australian researchers,
Lorell Burton and Gerald Fogarty, provides
a good idea of the nature of visual-spatial
ability.49 Previous research on the three-
stratum model had identified the following
specialized factors for visual-spatial abilities.
Closure of forms (CF): The ability to
detect a hidden form in a larger display,
particularly if the detection requires that
the examinee “break a set” of looking at a
larger or different figure. As an example,
count the number of vertical lines in this
sentence. To do so you have to detect
features within familiar letters.
Speed of closure (CS): The speed
with which you can recognize easily
detectable forms, such as a triangle or
circle. This is closely related to the more
general concept of processing speed.
Speed of rotation (SR): The speed with
which one can recognize a simple figure
(e.g., a letter) rotated out of its usual
orientation.
Visualization (Vz): The ability to
imagine what a visual figure will look like
48 Carroll, 1993; Horn & Stankov, 1982.
49 Burton & Fogarty, 2003.
THE MECHANICS OF INTELLIGENCE
167
Table 6.2. Loadings of visual-spatial and imagery first-order factors on a general
visual-spatial reasoning factor
CF CS SR VZ Image Quality Image Speed Image Self-report
Loading .74 .69 .80 .87 .83 .59 .46
Source: Data excerpted from Burton & Fogarty, 2003, with permission from Elsevier.
if its orientation is changed. An example
is the “paper folding” test, in which the
examinee is shown a flat piece of paper
with dotted lines on it. The examinee
is to indicate what figure can be con¬
structed from folding the paper along the
dotted lines.
Memory for shapes (MS): The ability
to hold a shape in short-term memory
for a brief while.
In the terms of the g-VPR model
these abilities would be second-stratum
factors - more general than specific tasks
but still highly specific abilities. They would
be subsumed by the more general P and
R dimensions at the third stratum of the
model.
Tests of the abilities just listed, and a
number of others, were given to slightly over
200 undergraduate students [approximately
evenly split between men and women).
Table 6.1 shows the resulting correlation
matrix. The correlations are substantial and
positive, an indication of a single factor,
visual-spatial ability.
6.5.1. Imagery
Visual-spatial reasoning tasks force exami¬
nees to think about things that they can
see. Studies of imagery require thinking
about things that the examinees imagine.
For instance, you could ask a person to imag¬
ine a letter E in its normal form, and then
ask him or her to describe how it would
look if it were to be rotated ninety degrees
clockwise. An extensive body of research has
shown that the sorts of operations that can
be performed on images are at least loose
analogs to the sorts of operations, like rota¬
tion, that can be applied to a percept.50 If
50 Kosslyn, 1980.
operations on perceived and imagined fig¬
ures use the same mental operations, indi¬
vidual differences in visual-spatial ability
should be related to individual differences
in imagery.
Burton and Fogarty included a number
of objectively measured imaging tasks in
their test battery. An example of such a
task is given in panel 6.8. Performance on
images reduced to two factors: image qual¬
ity, the accuracy of information contained
in a person’s image, and imaging speed,
how fast a person can construct an image.
Table 6.2 shows the loadings of both percep¬
tual and imagery factors on a single visual-
spatial reasoning factor. Clearly imagery and
visual-spatial reasoning are closely related.
Note, though, that people’s reports of their
imagery have a lower relation to the visual-
spatial factor than the behavioral measures
of imagery do. This is interesting, because
many studies of imagery have relied on self-
report, which may not be terribly accurate.
6.5.2. Spatial Orientation
Spatial orientation is the ability to develop
an internal representation of an exterior
space, including one’s own position in that
space. There are very large individual dif¬
ferences in people’s ability to do so. Some
individuals seem to have a keen awareness of
their spatial orientation, both with respect
to objects in their immediate environment
[e.g., what is immediately to your right
rear?) and with respect to larger spatial lay¬
outs (e.g., can you draw the layout of the
building you are in right now?). Others have
a great deal of trouble answering these sorts
of problems. This does not mean that these
people go around lost! Instead they seem to
rely on memories for routes between key
locations. Panel 6.9 describes a case that
i68
HUMAN INTELLIGENCE
Panel 6.8. An Imagery Task
Here is an example of a creative imagery
task. The instructions are
Imagine the letter A. (Pause) Now imag¬
ine a triangle to the right of the A. (Pause)
Imagine this triangle so that it is now
facing upside down. (Pause) I want you to
place the letter A inside the center of the
triangle such that all end points or edges
match up.
Participants were instructed to write
down as many of the emergent forms as
they were able to detect. They were told
not to draw the image held in their mind
until they had written down everything
that they could see in their image.
A picture of what a participant might
image is shown immediately below. The
emergent forms detected might include
one small triangle, two larger triangles, a
diamond shape, the letter “w,” and so on.
Panel 6.9. Lost in the Hospital
A study was conducted of the extent
to which nurses understood the layout
of a hospital in which they had worked
for over a year. All the nurses regularly
went about their daily activities with¬
out getting lost. The experimenter asked
the nurses how they would go from one
familiar location to another if their nor¬
mal route was blocked. Some of the
nurses could develop alternate routes;
some could not.*
This is an example of a more general
finding. When people explore a space,
some of them will develop a mental rep¬
resentation functionally equivalent to a
mental map of the environment. This is
sometimes referred to as a “surveyor’s
representation.” Others do not reach this
level of understanding of their surround¬
ings. Instead they develop memories for
routes from one location to another. f
* Moeser, 1988.
1 Hunt, 2002, Chapter 6, section 5.
illustrates, dramatically, how strong these
individual differences can be.
A number of laboratory tasks have been
designed to evaluate different aspects of ori¬
entation ability. This research has generally
been conducted outside of the mainstream
of research on intelligence, even though
maintaining orientation is certainly part of
the concept of intelligence.
The reason for this situation is under¬
standable. In order to evaluate someone’s
orientation ability you have to determine
how well they can explore an unfamiliar
environment. This is an expensive, time-
consuming process that does not fit into
the classic testing paradigm. About the only
way that orientation can be evaluated under
normal testing conditions is to show the
examinee one or two pictures of an object,
taken from different perspectives, and ask
what the object will look like from another
perspective. This gets at one aspect of ori¬
entation, but only one.
Virtual environment technologies pro¬
vide a way of getting around the logisti¬
cal problems inherent in evaluating spatial
orientation. 51 In this technology the exam¬
inee has to move through computer¬
generated artificial worlds. Large individ¬
ual differences in orientation are found.
These are related to, although not identical
to, both conventional tests of visual-spatial
ability, particularly spatial rotation and
51 Waller, Beall, & Loomis, 2004.
THE MECHANICS OF INTELLIGENCE
169
perspective-taking task s, and skill at orien¬
tation in real-world environments.52
Spatial orientation serves as a shining
example of a cognitive skill that is socially
important for which there are wide indi¬
vidual differences, and, as we shall see in
the following chapters, that has both bio¬
logical and cultural bases, but that has been
almost ignored in research on intelligence
due to overreliance on the standard testing
paradigm.
6.5.3. The Relation between
Visual-spatial Reasoning and
Information Processing
How do information-processing models
enhance our understanding of visual-spatial
reasoning? The answer to this turns out, sur¬
prisingly, to be “very little,” not because
information-processing models are irrele¬
vant, but because they are incorporated into
several of the psychometric tests used to
identify visual-spatial processing. This can
be seen by contrasting two workhorse tasks
used to define verbal intelligence - vocab¬
ulary tests and paragraph comprehension
tests - to two workhorse tasks used to define
visual-spatial reasoning - closure and rota¬
tion tasks.
The two verbal reasoning tasks are com¬
plex in themselves. Recognizing the mean¬
ing of a string of letters requires identifica¬
tion of the string as a word, retrieving its
several meanings, and determining the cor¬
rect meaning in context. The selection pro¬
cess is not trivial. In English, which appears
to be a particularly ambiguous language,
the typical word has 2.5 meanings.55 Para¬
graph comprehension requires recognition
of words, retrieval and selection of word
meanings, plus analysis of sentences, and
then the construction of text and situa¬
tion models. Neither psychometric task goes
to the level of detail in examining verbal
processing that the information processing
measures do.
52 Hegarty & Waller, 2005; Waller, Knapp, & Hunt,
2001.
53 Hunt & Agnoli, 1991.
Figure 6.4. A visual closure task. There are five
topi (a species of antelope) in this picture. Can
you find them? (Ngorongoro crater, Tanzania,
August 2007. Photograph by the author.)
Closure tasks require the detection of
lines, assignment of boundaries to objects,
detection of surfaces, and development of
representations of objects from represen¬
tations of surfaces. There is something of
an analogy to building a text model, or
even a situation model. The example in
Figure 6.4 illustrates this; perception is dif¬
ficult until you (a) have a clue about what
you might see and [b) are given an anchor¬
ing point. This level of detail is close to
that used in experimental studies of figure
detection.
In a rotation task a person is shown a
relatively simple picture and asked what it
would look like if it were to be presented at a
different orientation. Figure 6.5 presents an
example. There are strong individual differ¬
ences in the ability to deal with tasks such
Figure 6.5. A rotation problem. If gear A is
moved in the direction indicated by the arrow,
which way will gear C move?
HUMAN INTELLIGENCE
170
as this.54 Rotation tasks actually appeared
on psychometric tests of visual-spatial ori¬
entation before they were studied in exper¬
imental laboratories. The only difference
between the psychometric and information¬
processing techniques is that the psycho¬
metric tests determine how many rotation
problems a person can solve in a fixed period
of time; while the laboratory procedures
determine how long a person takes to solve
individual rotation problems.
Visual-spatial reasoning factors generally
have high loadings on the general reasoning
factor, g. From an information-processing
view this is not surprising. Visual-spatial
reasoning tasks often require the develop¬
ment and manipulation of internal rep¬
resentations of visual-spatial information.
Baddeley's model of working memory con¬
tained a "visual -spatial scratch pad." Subse¬
quent research has shown that measures of
Baddeley's scratch pad and operations on its
content are apart from, although correlated
with, measures of verbal working memory.55
As was the case in discussing the work¬
ing memory-g relation, trying to pick visual-
spatial working memory apart into sepa¬
rate tasks does not reveal any single, crucial
process that explains visual-spatial reason¬
ing. The visual-spatial working memory sys¬
tem is just that, a system. Reasoning about
percepts, images, and space is an emergent
property produced by the synthesis of stor¬
age, attention to, and processing of, percepts
and images.
6.6. Summary Comments on
Information Processing and
Intelligence
Individual differences in information¬
processing capacity make a substantial
contribution to individual differences in
cognitive skills - in essence, to intelligence.
However, the two are not identical. Rather,
information-processing capacity constrains
a person's intelligence.
54 Hegarty, Just, & Morrison, 1988.
55 Ackerman, Beier, & Boyle, 2002; Logie, 1995.
Thought results from the development
of internal representations of external prob¬
lems. This means that there has to be a stor¬
age place (or places) for the representation
of percepts and images. This includes the
possibility that there are separate storage
places for, say, linguistic and visual-spatial
information. There have to be processes
for fetching information and transforming
it, in an orderly fashion. There has to be
some way of prioritizing the processing of
crucial information. That is what the con¬
trol of attention is all about. Everything has
to work together. And finally, it is better
if everything is done quickly. The impor¬
tant question to ask about the information¬
processing contribution to intelligence is not
whether this or that isolated process is effi¬
cient or deficient. It is whether the total sys¬
tem is functioning well enough, and quickly
enough, to get the job done.
Information-processing capacities alone
do not produce intelligent behavior. Intel¬
ligent behavior is evidenced by appropriate
responses to external problems. Choosing
the right response requires the construction
and manipulation of an internal representa¬
tion of a problem. Take an illustrative case,
buying a new car.
In order to decide what car to buy you
have to think about the different ways that
you are going to use the vehicle, the capabili¬
ties of automobiles offered for sale, the price
range, the cost of operation, and so on and
so forth. Working memory is a shorthand for
the mental workspace and tools you use to
build and manipulate your representation of
the situation. The more pieces of informa¬
tion you can consider at once, the better
you can balance relevant information. The
faster your processing speed, the less likely
it is that information in the transient work¬
ing memory stores will decay, and the more
likely it is that you will be able to juxtapose
relevant pieces of information. For instance,
you must realize that economic considera¬
tions include both the purchase price and
the costs of operation and insurance.
I could just as easily have taken an exam¬
ple from cooking, political decision mak¬
ing, computer programming, or any other
THE MECHANICS OF INTELLIGENCE
field where people use their ability to think.
Thinking about anything requires having rel¬
evant knowledge. In order to bring that
knowledge to bear you have to use the
machinery provided by your information¬
processing system.
Although no one of the components
of the information-processing system dic¬
tates thought, each of the components
constrains thought. Just how different con¬
straints operate depends on both the situa¬
tion and the person. We have seen exam¬
ples of situational variation throughout this
book. The Wonderlic Personnel Test [WPT,
see Chapter 2) evaluates a person’s ability
to answer many simple questions, quickly.
Processing speed and the ability to switch
tasks (control of attention, executive func¬
tioning) constrain performance on the WPT.
Now recall the convoluted sentence on lib¬
eralism written by the Dean of Yale’s Law
School. It took a big working memory to get
through those forty-five words.
The nature of information-processing
constraints varies in different populations.
Comparing results from studies of intelli¬
gence in university populations and in the
elderly, we find that the variations in pro¬
cessing speed are not major determinants of
performance for sophomores, but they are
for senior citizens. Sophomores are pretty
good at thinking quickly. One of the most
consistent findings in research on aging is
that processing speed decreases substantially
over the adult years, possibly beginning in
the thirties.56
56 Salthouse, 1996.
171
We can see a similar effect in studies of
the role of attention, short-term memory
storage, and information-processing mea¬
sures in general. Studies that focus on intel¬
lectually capable young adults are examin¬
ing a group in which individual differences
in constraints due to information-processing
capacities are small, relative to their role
in other populations. Attentional control is
weak in young children and the elderly; pro¬
cessing speed decreases markedly with age,
even during an adult’s working lifetime; and
information-processing constraints appear
to be much stronger in people of low intel¬
ligence than in those with normal or high
intelligence, as assessed by test scores.57 The
relationships among these findings have not
been ignored, but they certainly have not
been studied adequately. In a review of stud¬
ies of the relation between working mem¬
ory and intelligence, Ackerman and his col¬
leagues observed that less than three percent
of the correlations reported involved partic¬
ipants over thirty years old!58
Psychologists would do well to consider
this fact:
All sophomores are human. Not all
humans are sophomores.
Future research on the relation between
intelligence and information processing
ought to ask how information-processing
capacities constrain intelligence in differ¬
ent populations, as well as how information
processing constrains thought in people in
general.
57 Detterman & Daniel, 1989.
58 Ackerman, Beier & Boyle, 2005.
CHAPTER 7
Intelligence and the Brain
The human brain, a 3-pound mass of
interwoven nerve cells that controls our
activity, is one of the most magnificent -
and mysterious - wonders of creation.
President George H. W. Bush,
July 17, 1990 (Presidential
Proclamation 6158 designating the
1990s as the “Decade of the Brain”)
He's a nice guy but he played too much
football with his helmet off.
President Lyndon Johnson,
referring to Congressman (later
President) Gerald Ford. Ford had
been a collegiate football star.
Attribution by Schnakenberg
(2004) .
The two presidents were right. President
Bush (or his speechwriter) described the
brain accurately — a very complex bit of
circuitry. President Johnson’s description of
Ford may not have been accurate, but he
was right that bouncing the brain is not a
good thing. A modem playwright put the
matter in a slightly different way.
Merkin's brain has a mind of its own.
Act I of Below the Belt, a 1997 play
by Richard Dresser
Every expression of intelligence is due to
actions of the brain. What actions a brain
will take in a given situation depends upon
both the brain’s structure and its history.
This chapter will focus on individual dif¬
ferences in brain structures and processes
related to intelligence. The topic is exciting.
Findings are coming in so fast that it is hard
to make sense of all of them. My computer
literature search for papers on BRAIN AND
INTELLIGENCE retrieved 5,648 citations.1
Neither I nor anyone else has read them all.
There are some who believe that virtually
all psychology, including the study of intelli¬
gence, will soon be reduced to studies of the
brain. I do not think so. What I do think,
and what will be stressed throughout this
1 Using PsycINFO, 30 May, 2009.
1J2
INTELLIGENCE AND THE BRAIN
m
chapter, is that the relation between intelli¬
gence and the brain is very important, but
has to be kept in perspective. Consider this
analogy.
Orchestras vary greatly in their ability to
play music. The same piece played by your
local high school orchestra and by the New
York Philharmonic may be recognizably the
same piece, but the two orchestras do not
sound the same. There is a positive correla¬
tion between size and musical quality. High
school orchestras tend to be smaller than
citywide amateur and professional orches¬
tras, which are in turn smaller than the
major philharmonics. However, there can
be a good deal of variation in the quality of
music performed by orchestras of the same
size. Adding performers to your high school
orchestra will not make it sound like a phil¬
harmonic.
There is also a correlation between intel¬
ligence and brain size, but the same cautions
apply.
By analogy to studies of brain injury,
imagine studying the essence of orchestral
quality by removing one player at a time. To
start, remove the conductor. The orchestra
begins to play in a flat, hesitant fashion, and
can play only those pieces for which it has a
lot of practice. But it can play, and besides,
there are inter-orchestra differences when
the conductor is there. In addition, there
is the puzzling problem that the conductor
clearly influences the music being played, in
spite of the fact that the conductor does not
make any noise!
Now try removing all or part of the
string section, the brass, the woodwinds, or
the percussion instruments. Each removal
would affect performance, but the effects
would depend upon the work being played.
Removing the strings would affect almost,
but not quite, all pieces. Playing “76 Trom¬
bones” does not require a violin; percussion
instruments are not needed to play chamber
music. And there is still the problem that
there is great variation in the performance
of intact orchestras.
Now try an analogy to modern studies of
how brain metabolism varies with mental
activity. The orchestral equivalent would be
measuring the sound level in different parts
of the orchestra. You would quickly find
that in all but the simplest pieces the sound
comes from all over, and that the pattern of
sound varies far more with the piece being
played than with the quality of the orches¬
tra. The same thing is true in cognition; all
but the simplest problem-solving activities
elicit neural activation across the brain, and
the pattern of activation varies more with
the activity than with the individual.
Instead of looking at activity all over the
orchestra, suppose that we arrange a "labora¬
tory study” that isolates the performance of
individual players, so that we can rate their
performance. This is what the information¬
processing psychologist does, by designing
situations that isolate working memory or
visual perception, instead of having them
work together, as they do in everyday prob¬
lem solving. You would find that you were
getting somewhere, for there would be a
substantial correlation between the qual¬
ity of individual performers and the quality
of the orchestra. Musicians in philharmon¬
ics are much better than musicians in high
school orchestras. However, if you were to
look at a narrower range - say, between
the members of the New York, Cleveland,
Chicago, and Seattle Philharmonics - you
would find that the differences were very
small.
Then there is the pesky problem of the
conductor. You would find it very hard to
evaluate a conductor without an orchestra.
He or she would look more like a person
suffering from a minor epileptic fit than a
musician, and would not make any sound at
all. How can you reconcile this with the fact
that that orchestras play considerably better
with a conductor than without?
In desperation you conduct a "metabolic”
study, by looking at how much orchestras
are paid. This is an indicator. Musicians
in major orchestras earn considerably more
than musicians in minor ones, and amateurs
are not paid at all. But is this because being
paid more produces better music, or is it
perhaps the other way around?
Besides, the silent conductor is getting
paid more than anyone.
*74
HUMAN INTELLIGENCE
The brain is more complex than an
orchestra, and cognition is more compli¬
cated than musical performance. Cognition
and music are emergent properties. They
depend partly upon measurable qualities of
the parts of the organisms that produce
them, brain and orchestra, and partly upon
the interaction between those parts. We
can make progress in understanding cogni¬
tion [and music) by making measurements
on parts, but these measurements, alone,
will not provide a full explanation. You
should keep this in mind as we discuss the
brain-mind relations that tell us a lot, but
not the whole story, about intelligence.2
The following two sections provide an
introductory discussion of the structure of
the brain and of modern technologies for
examining brain structures and processes.
Readers familiar with both topics can jump
immediately to section 7.3, which begins
the discussion of major findings relating
brain variables to intelligence. I urge readers
not familiar with brain structure or the new
technologies to refrain from jumping too
quickly.
7.1. The Structure of the Brain
The human brain is a swelling that sits at
the upper end of the spinal cord. Figure 7.1
presents a “cartoon” version of the brain, as
viewed from the left. It is divided into four
major anatomical structures, called lobes -
the frontal, temporal, parietal, and occipi¬
tal lobes. The occipital lobe, at the back of
the brain, is primarily concerned with visual
analysis. It also plays a role in visual reason¬
ing. The cerebellum, sitting below and to
the rear of the cerebral cortex, is largely con¬
cerned with automatic motor coordination,
although it does have some function in
cognition.
2 I am not arguing for a duality of mind and brain. If
we knew the nature of every connection between
the approximately five billion neurons in a person’s
brain, and if we knew the algorithms the brain uses
to activate and alter these connections, we would
know everything there is to know about that per¬
son’s cognition. We are so far from having such
knowledge that, for the foreseeable future, there
is a place for nonbiological models of intelligence.
Figure 7.1. A sketch of the human brain, seen
from the left. Sketch by the author.
If you view the brain from above you
would see that a deep fissure divides it into
left and right hemispheres, connected by
neural bundles that bridge the fissure. The
largest of these bundles is the corpus callo¬
sum, which provides the main communica¬
tion link between the two hemispheres.
In general, somatosensory and motor
information is represented contralaterally;
the right side of the brain controls the left
side of the body, and vice versa. There are
some exceptions, notably in the analysis of
information in the middle of the visual field,
but these need not concern us. Certain func¬
tions that are not directly tied to a side of
the body may also be differentially localized
across the hemispheres. Language, which
primarily depends on structures in the left
hemisphere, is the best known of the later-
alized functions.
There are individual differences in brain
structures. Some left-handed people have
their language centers located in the right
hemisphere, and there are some differences
between men and women in the localization
of brain functions. These will be discussed in
Chapter 11, section 3, which describes male-
female differences in intelligence.
Figure 7.2 shows a diagram of sev¬
eral subcortical structures that are impor¬
tant in cognition. The cingulate gyrus func¬
tions as a communication system between
various areas of the cerebral cortex. It
also appears to be important in response
selection. Below the cingulate gyrus is the
limbic system, which contains the hippocam¬
pus, and the amygdala. The hippocampus
INTELLIGENCE AND THE BRAIN
*75
Cortex
Figure 7.2. The cingulate gyrus and the limbic system. Drawing
courtesy of the National Institutes of Health.
plays an important role both in memory and
in spatial-visual reasoning. The amygdala,
along with a midbrain structure called the
hypothalamus, is involved with emotions.
Six terms are used to define locations in
the brain. They are:
Frontal: The forward part of the brain or
part of the brain being discussed.
Posterior: The opposite of frontal.
Ventral: Toward the bottom of the brain
or the brain region being discussed.
Dorsal: Toward the top of the brain or
brain region.
Medial: Toward the middle of the brain.
Lateral: Toward the side of the brain.
These terms may be combined. For
instance, the dor so -lateral prefrontal cortex
(DLPFC) is an area at the extreme front of
the brain (prefrontal], on the upper (dorso]
outside (lateral] surface. It is important in
working memory.
All information held in the brain (mem¬
ories] and all processing upon that informa¬
tion (thoughts, percepts, images] is deter¬
mined by the physical form of networks of
nerve cells, or neurons. The term “physical”
is important; ultimately all thinking comes
down to the manipulation of neurons. So is
the term “network”; we store and process
information by changing the configurations
of networks of connected neurons, not by
changing the states of individual neurons.
Our memories of our grandmother, or of
yellow Volkswagens, are coded as configu¬
rations of neural elements. Within limits,
two different individuals may code the same
information in somewhat different neural
configurations.
Neurons can be classified, anatomically,
as either gray matter or white matter. White
matter refers to neurons whose axons are
coated with a fatty substance called myelin.
Gray matter consists of masses of neurons
that are involved in computations within a
local area of the brain. The white matter
provides long-distance connections between
brain regions. There is a loose analogy here
to distributed computing, where networks
of gray matter play the roles of local comput¬
ing centers and the white matter provides
the cabling that connects the local centers
to each other.
Introductory textbooks often contain
maps showing where different functions lie
in the brain. The most famous examples
involve speech. Broca's area, in the left
posterior frontal region, is associated with
speech expression, and Wernicke's area , in
the left temporal lobe, is associated with
retrieval and understanding of semantic and
176
HUMAN INTELLIGENCE
syntactic relationships. A great deal has also
been written about how the left brain is
specialized for analytic, sequential reasoning
while the right brain conducts intuitive, par¬
allel reasoning. This is an oversimplification.
The brain does contain centers that are
specialized for certain kinds of processing,
but it does not contain regions dedicated
to broad cognitive or emotional functions.
The brain functions as a system. The orches¬
tral analogy is apt. In understanding human
intelligence it is useful to consider what cog¬
nitive functions are carried out in different
regions of the brain, but understanding as
complex a behavioral phenomenon as intel¬
ligence requires understanding both where
different specialized processes are and how
they mesh to produce thinking.
7.2. Technologies for Examining
the Living Brain
Until the discovery of X-rays at the turn
of the twentieth century our knowledge
of brain structure was based upon post
mortem examinations. These provide a lim¬
ited source of information about brain-mind
relations, because post mortem examina¬
tions are biased toward studies of the elderly
or those who have died an untimely death.
In both cases the state of the brain fol¬
lowing death may not accurately reflect its
state when the person was alive and fully
functioning. Nevertheless, a great deal of
information about brain-mind relations was
obtained in the nineteenth and first three
quarters of the twentieth century by study¬
ing alterations in behavior following brain
damage.
In one of the pioneering studies in this
field, the nineteenth-century physician Paul
Broca (1824-1880) determined that injuries in
the left posterior frontal lobe are associated
with deficiencies in speech expression but
not in speech comprehension. This condi¬
tion is referred to as aphasia. A century later,
in 1957, the world saw a striking example of
the malady. President Dwight Eisenhower
developed a minor aphasia, probably due to
a small aneurism in Broca's area. Eisenhower
recovered sufficiently to complete his term
of office (until January 1961), although he
continued to exhibit signs of mild speech
impairment. Given the extremely demand¬
ing nature of the presidency, Eisenhower’s
performance provides a striking example of
the dissociation between speech production
and comprehension.3
Broca's observations began a long and
fruitful line of research in neuropsychology ,
the study of the relationship between the
brain and mental functions. Neuropsycho¬
logical findings are highly relevant to the
study of intelligence, but they are limited in
an important way. Variations in intelligence
within the normal range may not be pro¬
duced by the same mechanisms that make
it possible to perform a particular cognitive
function in the first place. Consider the fol¬
lowing analogy. Loss of a leg makes a dra¬
matic difference in running speed. “Num¬
ber of legs” is not a determining factor in
running speed within the normal range of
variation.
In order to go beyond neuropsychology
scientists needed some method of examin¬
ing the brains of healthy individuals. X-ray
imaging, developed early in the twentieth
century, was a beginning; but early X-ray
images could not provide a clear picture of
the soft tissue in the brain. In the 1970s much
more powerful imaging techniques became
available. They fall into two broad classes:
techniques for measuring brain structures
and techniques for measuring neural activ¬
ity at different sites in the brain.4 Panel 7.1
presents a brief, nontechnical description of
the major technologies in use today. The
development of new technologies contin¬
ues, so my list will probably be outdated
within a year of the publication of this book!
The results from studies using these tech¬
niques are exciting, but we want to keep
three recurring problems in mind.
3 Eisenhower recovered. Other leaders have not been
so fortunate. In 1526 King Henry VIII of England
received a blow on the head while jousting. His
impulsive behavior following the injury, which dis¬
rupted both his personal life and English policy,
suggests forebrain damage.
4 Haier, 2009.
INTELLIGENCE AND THE BRAIN
177
Panel 7.1. Technologies for Looking
at the Brain
This panel presents a brief view of
the major technologies used, as of 2010,
to relate brain structures and processes
to intelligence. Each of these technolo¬
gies was developed from basic discover¬
ies in physics that made it possible to
detect weak electrical and magnetic sig¬
nals emanating from the brain. Making
sense of the signals required the devel¬
opment of complicated computer algo¬
rithms and extremely rapid electronic
computing machinery, whose develop¬
ment also required basic advances in
physics. The imaging technologies rep¬
resent a striking example of how basic
research in one field can have important
practical implications in other fields.
Technologies for Examining Brain
Structures
Computerized Axial Tomography
(CAT or CT Scanning)
CT scanning is derived Irom X-ray med¬
ical imaging. Marie Curie received the
Nobel Prize for her discovery of X-rays in
1911. This started a string of Nobel awards
building on Curie’s discovery. The first
CT scanning devices were announced in
1972, and the inventors were awarded
Nobel Prizes in 1979.
In CT scanning, low-intensity X-rays
are passed through the body from posi¬
tions on a circular ring around the body.
This contrasts with conventional X-ray
imaging, in which a single picture is taken
by passing X-rays through the body onto
a photographic plate. After the X-rays
have passed through the body they are
collected by receivers that are far more
sensitive than the chemical elements on a
photographic plate. This permits the use
of X-rays of low enough intensity that
their passage is impeded by soft tissue,
producing a soh-tissue image. The two-
dimensional images, taken from many
different angles, are combined by a com¬
puter program that determines the struc¬
ture of various parts of the brain. The
method is applicable to all parts of the
body, not just to the brain.
Magnetic Resonance Imaging (MRI)
Magnetic resonance imaging (MRI)
sprang from work in the 1960s and 1970s,
based on discoveries about electromag¬
netism that date from the 1930s and 1940s,
including Nobel Prize-winning research
by Isidor Rabi. The first medical scan¬
ners were introduced in the early 1980s.
The developers of modern MRI, Paul
Laterbauer of the University of Illinois
and Peter Mansfield of the University of
Nottingham, were awarded the Nobel
Prize in 2003.
In MRI, the person being scanned is
placed in a tube that is surrounded by
a large magnetic field. A radio pulse is
then directed toward the part of the body
being scanned. The pulse frequency is
chosen to resonate with hydrogen atoms.
This causes hydrogen atoms in the body
to move out of alignment with the mag¬
netic field. They then return to alignment
and, as they do so, emit a detectable elec¬
tromagnetic signal. This information is
used to reconstruct locations in the brain
or body.
Diffusion Tensor Imaging (DTI)
Diffusion tensor imaging (DTI) is a form
of magnetic resonance imaging based on
signals that are sensitive to the diffu¬
sion of water molecules along the myelin
sheaths (white matter) that coat some
of the neurons in the brain. Signals are
transmitted along myelinated neurons
considerably faster than they are along
(continued)
178
HUMAN INTELLIGENCE
Panel 7.1 (continued)
unmyelinated neurons (gray matter).
Columns of myelinated neurons are
believed to be involved in transmitting
signals from one region of the brain to
another, while unmyelinated neurons are
involved in local computations within a
region. As a loose analogy, DTI provides
a way of locating the cabling between the
computing centers of the brain.
Technologies Used to Identify
Brain Activity
Positron Emission Tomography (PET)
Scanning
The basic ideas behind positoron emis¬
sion tomography were developed in the
1950s. The first use in humans waited until
the 1970s, due in part to the need for high¬
speed computing to support the sensing
technology.
As radioisotopes decay they will
emit positrons (anti-electrons). When a
positron encounters an electron both are
annihilated, emitting a gamma ray. In
PET scanning a person is placed in a
ring of sensitive sensors. A rapidly decay¬
ing radioisotope is then injected into
the bloodstream. Rapidly decaying iso¬
topes are required in order to avoid dam¬
age to tissue as the gamma rays pass
through the body. The isotope will be
taken up by tissue, roughly in propor¬
tion to the metabolic activity in neu¬
rons at that point. As the isotope decays
it emits positrons, which are annihilated
when they encounter electrons, produc¬
ing gamma rays. The gamma rays will
vary in strength according to tissue den¬
sity and current metabolic activity at that
point. Sensors detect the gamma rays as
they exit the body. Computer programs
are then used to calculate metabolic
activity at various locations.
Functional Magnetic Resonance
Imaging (fM RI)
Like MRI, fMRI is based on molecules
generating an electromagnetic pulse in
the presence of a magnetic field. The
biological mechanism is different. When
neurons are active they take up oxygen
from the blood. This causes a change in
the magnetism of the oxygen molecules,
which can be detected by sensors.
The signal is called the Blood Oxygen
Level Dependent (BOLD) response. The
BOLD response provides an indication of
neural activity at a location.
Electroencephalograms (EEG) and
Event Related Potentials (ERP)
Neural events generate electrical signals
that can be detected by electrodes placed
on the scalp. This was first demonstrated
in 1929. It is the basis for the modem elec¬
troencephalogram (EEG). EEG record¬
ing is important in medicine because
physiological states have characteristic
EEG signatures. These include character¬
istic patterns for epilepsy and for different
stages of sleep. The event-related poten¬
tial (ERP) is an EEG signal in response
to a specific external event, such as a
flashing light. The ERP has proven to
be a very useful tool in cognitive psy¬
chology, as different mental events have
characteristic ERP signals. For example,
when people hear semantically mean¬
ingless sentences, such as THE COOK
ROASTED THE CEMENT, they dis¬
play a characteristic ERP. Syntactically
anomalous sentences, such as WOMAN
THE LOVE CATS, have a different ERP
signature.
Modern EEG signals are recorded from
50 to more than 100 sites on the scalp.
Computer-based analysis is required to
identify signature waveforms and to inter
their location in the brain.
INTELLIGENCE AND THE BRAIN
1 79
7.2.1. The Spatial and Temporal
Resolution Problem
The MRI, PET, and CAT procedures descri¬
bed in panel 7.1 are accurate to within a
few millimeters. Therefore, they can detect
structures and processing activity in bun¬
dles of neurons, but not individual neu¬
rons. Timing presents a problem. Func¬
tional MRI (fMRI), the most widely used
method, identifies neural activity by mea¬
suring oxygen uptake, referred to as the
BOLD response. The BOLD response takes
place over a few seconds. By contrast, cog¬
nitive operations such as word identification
take place in well less than a second. There¬
fore, the BOLD response is useful in locat¬
ing a place of brain action, but less useful in
isolating the time course of the action.
There is activity all over the living brain,
all of the time. In order to determine the
place of action associated with a particu¬
lar cognitive activity neuroscientists have to
isolate the “signal," the brain activity asso¬
ciated with the cognitive action of interest,
from the “noise,” the background activity of
the brain. In order to do this neuroscientsts
use an analog of the Donders subtraction
paradigm described in Chapter 6. The brain
regions supporting different steps within a
cognitive activity are located by compar¬
ing the BOLD response during an activity
that employs the function of interest to the
BOLD response during a control condition,
which contains all the steps of the first activ¬
ity except the one of interest. For exam¬
ple, in order to identify the neural struc¬
tures associated with lexical identification
the BOLD response obtained while a per¬
son reads words like CAMEL is compared
to the BOLD response as he or she looks at
a meaningless letter string, such as LEMAC.
The BOLD response to LEMAC is sub¬
tracted, on a point-by-point basis, from the
BOLD response to CAMEL. The areas of
the brain that show more activity in response
to CAMEL than to LEMAC are assumed to
be involved in the retrieval of meaning. As
in the case of the use of Donders’s paradigm
in reaction time studies, the comparison is
valid only if the brain regions involved in
executing the function of interest are unin¬
fluenced by the execution of other functions
that may be active at about the same time.
EEG/ERP recordings of electrical events
are accurate to within a few milliseconds.
The neural response to a stimulus will fol¬
low presentation of a light or noise by a few
milliseconds, simply because there has to be
time for the signal to get to the brain before
it can be interpreted. It is difficult to deter¬
mine the place in the brain that is the source
of an EEG/ERP signal unless a large number
of electrodes are placed on the scalp. Large
electrode arrays did not become practical
until the advent of high-speed computers,
but this is no longer a serious concern.
7.2.2. The Averaging Problem
All of these technologies rely on the analy¬
sis of signals immersed in noise. Therefore,
results are often reported in terms of the
average signal observed, where the averag¬
ing is over trials and sometimes over indi¬
viduals. A technology that may be sensi¬
tive enough to detect trends in averages may
not be sensitive enough to detect individual
differences around that trend. Accordingly,
someone interested in individual differences
in cognition - that is, intelligence - can
rely on positive results obtained using the
new technologies (subject, of course, to the
usual cautions about the need for replica¬
tion), but should be a bit cautious about
results that report no difference. Are such
results obtained because there really are no
differences between individuals or because
the technologies being used are not sensitive
enough to detect them?
Average patterns of brain activity, com¬
puted over individuals, may not represent
the pattern of activity in any one individ¬
ual. For instance, male and female patterns
of brain activity differ in a number of cog¬
nitive tasks. Therefore, it often does not
make sense to report average brain activity,
summed across men and women. This is a
glaringly obvious distinction, for we know
that there are many differences between
men and women, and so are alert to errors
based on averaging results across the sexes.
i8o
HUMAN INTELLIGENCE
What is more worrisome is that there may be
qualitatively different types of brain activi¬
ty within each gender that are masked in
averaged data.
7.2.3. Logistical Problems
All the new technologies except the EEG
rely on large, expensive, immobile machin¬
ery. This is not a problem that is likely
to be solved by technology, for the prob¬
lems are due to the physical nature of the
sensing apparatus. All the technologies use
machines that must be shielded from extra¬
neous electromagnetic fields. Magnetic res¬
onance imaging requires a large, inherently
heavy magnet.
Imaging experiments are time-consuming
and involve some physical discomfort for
participants. This can range from injections
of radioactive materials (in PET scanning] to
holding still in a noisy, uncomfortable cham¬
ber (in MRI). Therefore, there are pres¬
sures to use a small number of participants.
For instance, one important study in the
field reported correlations based on only
eight participants. Thirty to fifty participants
would be considered a large study. For sta¬
tistical reasons alone, positive results can be
trusted but negative results are indetermi¬
nate, due to the low statistical power of the
experiment. In addition, using very few par¬
ticipants, often drawn from available rather
than representative groups, raises the pos¬
sibility that there are important individual
differences in brain-behavior relations in the
population at large that are not represented
in the study group.
Because of the expense involved and the
need to study relatively few participants,
many studies are based on comparisons of
extreme groups - often a contrast between
“normals” (usually healthy, relatively young
adults] and clinical populations. While such
studies can be informative, they do exag¬
gerate the size of findings. They can also
miss important phenomena that appear only
when the full range of intelligence is studied.
The new imaging and EEG technologies
have provided types of data that the psycho¬
metricians of the twentieth century could
only speculate about. Findings based on the
new technologies have made, and will con¬
tinue to make, great strides in our under¬
standing of the biological basis of intelli¬
gence. Like any other technologies, these
new ways of looking at the brain do have
limits, and the limits should be kept in mind.
7.3. Brain Structures and Cognition
It would be nice if thinking were located
in neat compartments in the brain. As the
orchestra analogy suggests, this is a vain
hope. Consider an important part of cogni¬
tion, paragraph comprehension, that is eval¬
uated in almost all battery-type intelligence
tests.
From the viewpoint of a psychometri¬
cian, a test of paragraph comprehension pro¬
duces a score that is an indicator of verbal
intelligence. From the viewpoint of a cog¬
nitive psychologist or a neuroscientist, para¬
graph comprehension is staggeringly com¬
plex. It involves aspects of (at least] visual
form identification (occipital and temporal
lobes), control of eye movements (parietal
lobes], retrieval of lexical information (left
temporal lobe], retrieval of syntactic rules
(left temporal lobe), execution of syntacti¬
cal rules for sentence comprehension (left
frontal lobe), and maintenance of attention
on the topic (left frontal lobe, anterior cin¬
gulate cortex). It takes all of this, and more,
to read The Cat in the Hat to a four-year-old
child.
Nevertheless, we can make some gener¬
alizations about the functions of different
brain areas:
Frontal cortex : The frontal lobe is involved
in two major aspects of thought: the pro¬
vision of working memory and the control
of attention. Both working memory func¬
tions and the control of attention involve
coordination between the frontal and pari¬
etal lobes, and the anterior cingulate gyrus
(Figures 7.1 and 7.2). The frontal lobe is also
important in inhibition of action, includ¬
ing inhibition of emotion-guided responses.
Inhibiting such responses can be impor¬
tant, both in constructing plans and in
INTELLIGENCE AND THE BRAIN
181
Panel 7.2. The Case of Phineas Gage
Phineas Gage was a nineteenth-century
railway construction foreman. He was
known as a good supervisor and a shrewd
businessman. In 1848 an accidental explo¬
sion drove a steel rod through the for¬
ward part of Gage's skull. After he recov¬
ered he was irreverent, capricious, unable
to concentrate, and often profane. Some
reports remark on his tendency to make
sexual comments, something that was far
more proscribed in Victorian times than
it is today. Gage lost his job as a fore¬
man and became a stage coach driver.
This job probably minimized social inter¬
actions, compared to his earlier job, but
still required considerable skill. Ten years
after the accident he developed epilepsy.
He died in i860. During his lifetime there
was interest in his case. As late as the
lqqos, Gage's skull was on display in
Harvard’s Library of Medicine.
Over a hundred years after Gage’s
death modern neuroscientists examined
the skull and concluded that the iron bar
had heavily damaged the frontal lobe.*
Here is a personal anecdote. For some
years I supervised an undergraduate lab¬
oratory course in cognitive psychology.
One day when I entered the laboratory
I found that a graduate student leading
one of the sections, a small woman, had
been trapped in the corner of the room
by a large man who was berating her
for giving him low grades. His behavior
was threatening and completely inappro¬
priate. 1 intervened, and convinced him
to leave. (Because he also exhibited very
poor motor coordination I was sure that
I was in no physical danger.) I noticed
that he had scars on his forehead. Upon
later inquiry I learned that he had suf¬
fered frontal lobe damage.
These two examples illustrate the
impulsivity and inability to concentrate
that is typical of people with damaged
frontal lobes.
* Damasio et al., 1994.
maintaining social order. People with dam¬
age to the frontal lobes are notoriously
poor in both functions. Panel 7.2 presents
a famous illustrative case and a relevant per¬
sonal anecdote.
Temporal lobe: The temporal lobes are
heavily involved in hearing. The left tem¬
poral lobe and the left posterior frontal lobe
are important for the production and com¬
prehension of speech. There is also some
involvement of speech in the right tem¬
poral lobe. In about five percent of the
population the speech areas are found on
the right rather than the left. Such peo¬
ple are almost always left-hand dominant.
(The converse is not true. Many left-handers
have their speech center in the left hemi¬
sphere.) The posterior temporal lobe is also
involved in the identification of visual stim¬
uli, as explained in the following discussion
of the occipital lobe.
Parietal lobe: The parietal lobes were once
thought of as being responsible for analysis
of tactile signals and for coordination of
sensory-motor movements. These functions
are carried out contralaterally; the left side of
the brain receives sensations from and con¬
trols actions on the right side of the body,
and vice versa. We now know that the pari¬
etal lobe is also involved in the allocation
of attention to sensory input streams. In
addition, the parietal lobe plays an impor¬
tant role in locating objects and in sens¬
ing motion. It acts in conjunction with the
frontal lobe to support the working memory
system.
Occipital lobe: The occipital lobe is spe¬
cialized for visual analysis. This is where
most ‘low-level” visual analysis takes place,
including parsing the visual signal into con¬
nected surfaces and objects. During percep¬
tion visual information from the occipital
182
HUMAN INTELLIGENCE
Panel 7.3. The Case of H. M.
H. M. was a young Canadian architect’s
draughtsman who, in 1953, had his hip¬
pocampus, amygdala, and other parts of
the limbic system removed in order to
treat severe epileptic seizures. H. M. was
the last person on whom this operation
was ever performed. The surgery did con¬
trol his epilepsy, but it destroyed his abil¬
ity to form new declarative memories. As
a result he had to live in custodial care
until he died, in 2008, when he was in his
eighties. At that time his identity, Henry
Moulson, was revealed.
H. M. was studied extensively. One of
the most interesting aspects of his behav¬
ior was the highly selective nature of his
loss. After his operation his WAIS IQ
score was in the above-average range.
This is what would be expected, given
his employment prior to surgery. Declar¬
ative memory - that is, the sort of mem¬
ory that can be retrieved by an explicit
cue to recall - was gone. For example,
he could not recognize researchers who
had worked with him for over ten years.
However, he could learn new motor
skills, although he did not remember hav¬
ing learned them. He also showed some
semantic memory for events that had
occurred after his injury. Such memory
was reduced compared to the memories
of a noninjured person.
Although surgeons no longer perform
bilateral hippocampal removals, a few
cases of hippocampal loss due to injury
or accident have been reported. They all
resemble H. M.*
A good deal has been made of H. M.'s
above-normal IQ scores, which have
been used to argue that intelligence is
unaffected by the loss of declarative
memory. This conclusion defies logic. A
capacity for declarative memory is an
essential part of the concept of intelli¬
gence. Failure to evaluate it adequately
on the WAIS is a deficiency of the test.
You cannot say that a man who has to be
provided custodial care due to a cognitive
defect still has normal intelligence.
* Milner, 2005; Milner, Corkin, & Teuber, 1968.
lobe moves along a pathway through the
temporal lobes that is responsible for iden¬
tifying visual objects, and along a pathway
through the parietal lobes that determines
object location and movement.
Two subcortical structures are important
in cognition (Figure 7.2]:
The limbic system: The limbic system con¬
sists of the hippocampus, the amygdala, and
the fornix. The limbic system, and espe¬
cially the amygdala, is heavily involved in
emotional arousal, including fear. This is
important for cognition, because affective
reactions are important in capturing atten¬
tion and in response selection. One of the
important functions of the frontal and pre¬
frontal regions is the inhibition of reactions
based on affect, thus making cooler-headed
reasoning possible.
The hippocampus is central to the con¬
struction of declarative memory , memories
that can be recalled explicitly. People with
hippocampal injuries may exhibit a pro¬
found anterograde amnesia, in which they
are literally unable to form memories of
experiences following their injury. Panel 7.3
presents the case of H. M., which may
be the most-cited single case study in all
of psychology. The hippocampus is also
heavily involved in the learning of spatial
layouts.
Cingulate gyrus: This large structure lies
between the limbic system and the upper
parts of the cerebral cortex (Figure 7.2). The
cingulate cortex and the frontal system form
a circuit that is important for the control of
attention, in the sense of attending to cer¬
tain aspects of the current problem and not
INTELLIGENCE AND THE BRAIN
183
others. This is sometimes referred to as
"executive control.”5
This description of the functions of dif¬
ferent parts of the brain has not dealt with
individual differences. We now turn to a dis¬
cussion of how variations in brain structures
and processes are associated with individ¬
ual differences in cognition, that is, with
intelligence.
7.4. The Brain and General
Intelligence: g and Its Correlates
In a classic children’s story Winnie-the-Pooh
explains his mishaps by saying he is a "Bear
of very little brain.”6 Is it true that the big¬
ger our brains, the smarter we are? Panel 7.4
presents two cases in which men with large
brains displayed considerable intelligence.
But wait! The panel also contains a discus¬
sion of the problems of inferring scientific
laws from single case studies. The answer
to the question about brain size and intelli¬
gence depends on your perspective. Does a
large brain imply greater intelligence? From
an evolutionary point of view the answer is
"Definitely!” Within our species, the answer
is "Somewhat.” It also depends on what type
of thinking we are talking about and where
in the brain we look.
7.4.1. Evolutionary and Cross-species
Evidence
Cross-species and evolutionary records indi¬
cate a strong relationship between brain
size and cognitive power. Compared to the
brains of other mammals, the human brain
is relatively large, and very large for our body
size. Adult men have brains that are slightly
above 1400 cc5 in volume and 1.30 kilograms
in mass. Women have brains around eight
to ten percent smaller. Brains also vary with
age; they are obviously smaller in children,
and shrink in advanced age. Brain size is
positively related to body size, both within
and across species. A comparison to another
5 Posner et al., 2006.
6 Milne, 1926, Chapter 4.
large animal, the elephant, shows the human
advantage in brain size after body size is
considered.
The average weight of the elephant brain
(which varies depending on whether the
animal is male or female, and African or
Asian] is 4.70 kg, more than three times the
weight of the human brain. The elephant
weighs from 7,500 kg (African male) to 3,700
kg (Asian female). Modern humans weigh
around 75 kg (male) to 62 kg (female). Ele¬
phants are sixty to one hundred times heav¬
ier than humans, which makes the three-to-
one ratio of brain weights look a little less
important.
This example generalizes nicely. Across
species there is a remarkably accurate rela¬
tionship between brain size and body size,
E(Bi) = AgSir*,
where £(B[) is the predicted brain mass, in
grams, of the ith species, and S* is body mass
in grams. This is called an allometric equa¬
tion . The Ag and rg terms are constants that
depend upon the particular group of ani¬
mals, g, being compared. If the exponen¬
tial term, rgf is less than one, brain size is a
negatively accelerated function of body size,
which means that the increase in brain size
per unit weight of body size decreases as
the animal gets larger. The human:elephant
contrast illustrates this. The rg term is about
.75 across mammals, and .56 across birds.
Within a group of related species some
animals will have larger brains than oth¬
ers, after accounting for their body sizes.
The amount of variation that a particular
species displays, compared to others in the
same group, is indexed by the encephaliza-
tion ratio, Q, which is defined as the ratio of
the observed brain size to the size that would
be predicted on the basis of body size,
If Q is greater than one, the species has a
larger-than-expected brain size, calculated
by considering brain size:body size ratios in
comparable animals.
184
HUMAN INTELLIGENCE
Panel 7.4. Two Case Studies of Big
Brains, and a Comment on Case
Studies in General
Albert Einstein
The theoretical physicist Albert Einstein
(1879-1955) is often held up as the proto¬
typical example of high intelligence. Prior
to his death Einstein gave permission for
the scientific study of his brain. Einstein's
brain was not remarkable, for a man of his
age at time of death (seventy -six), except
in one way. His parietal cortex was about
15% larger than the brains of other men
who had died at about the same age.*
Both the scientists who conducted the
study and the popular press made a good
deal of this, observing that the parietal
cortex is involved in visual-spatial reason¬
ing and mathematical reasoning^ and that
Einstein had said that his thoughts tended
not to be verbal statements.
The First King of Scotland
Robert Bruce (Robert I of Scotland, 1279-
1324) had a spectacular career. In 1309 he
was supported by the Scottish church
as King of Scotland, even though he
had been excommunicated for murder¬
ing a political rival. In 1314 he defeated a
much larger English force at the Battle of
Bannockburn, establishing Scotland as an
independent kingdom.
A modern Scot, Ian Deary, used mag¬
netic resonance imaging (MRI) of a cast of
Bruce’s skull to estimate Bruce’s cranial
capacity.* He then determined the rela¬
tion between cranial capacity estimates
and intelligence, using a modern survey of
forty-eight adults, for whom he had both
an intelligence estimate (a reading com¬
prehension test) and measures of skull
size. Bruce had an estimated intracranial
volume of 1661.67 cc5; the mean of the
modern sample was 1492.50. Using the
capacity:brain size relation in the modern
sample as a guide, Deary and his col¬
leagues estimated Bruce's IQ to be 128,
approximately two standard deviations
above average. Deary concluded that
his estimate was consistent with Bruce’s
demonstrated skills as a political and mil¬
itary leader.
A Comment on Case Studies
Case studies make great reading. They
are often useful in suggesting hypotheses
about brain action, because they tell us
what variables are worth examining fur¬
ther. What case studies do not provide
is the sort of careful examination that
rules out alternative hypotheses. Con¬
sider what the cases of Phineas Gage
(panel 7.2), H. M. (panel 7.3), and Ein¬
stein and Robert Bruce contribute to sci¬
entific knowledge.
The case study of H. M. made sub¬
stantial contributions to our understand¬
ing of the mechanisms of memory. But
this case study was an unusual one. The
initial report of H. M.’s cognitive deficits
was followed by years of careful, well-
controlled studies of his cognitive capac¬
ities. Similar cases were sought out and
observed. The study of H. M. was valu¬
able because extensive observations were
possible, conceptual replications could be
made, and various alternative hypotheses
could be explored.
There was no systematic study of
Phineas Gage. It has been possible to
determine, to some extent, what his
injuries were, but we have only anecdo¬
tal evidence about his behavior. The evi¬
dence is moderately strong that he got
into more trouble after the accident than
before, but Gage’s behavior was never
recorded in anything like the detailed
records we have for H. M. However,
we do have well-documented records
of people whose injuries were similar
to those of Gage. Gage's story makes
INTELLIGENCE AND THE BRAIN
185
a nice illustration of points about brain
injury, but it is not scientific data.
The analysis of Einstein’s brain is inter¬
esting, but hardly conclusive. Many mea¬
surements were made comparing Ein¬
stein’s brain to the brains of lesser-known
people. If many measurements are made,
some of them are likely to meet con¬
ventional standards for statistical signif¬
icance by chance alone. The fact that
the difference was in the parietal lobe
is suggestive, for neuroscientific studies
have established that the parietal lobe
is involved in visual imagery and arith¬
metic. Einstein did claim that his internal
thought processes were visual rather than
verbal. However, self-reports of imagery
are not highly correlated with objective
indicators of imagery. §
Deary and colleagues’ estimate of
King Robert’s IQ is fun. Their conclu¬
sion is consistent with modern stud¬
ies of brain sizeiintelligence relations,
although Deary and colleagues’ estimate
of the IQ:cranial capacity size relation¬
ship was somewhat higher than is nor¬
mally found. While their estimate of
Bruce’s IQ was 128, the margin of error
was broad, with confidence intervals
from 106 to somewhere above 130. Most
national leaders probably have IQs in this
range. IQ estimates for the first forty-
two US presidents run from a high of
145-160 (Jefferson) to a low of 108-140
(Harding).**
And finally, whenever you deal with
case studies, you have to be concerned
about case studies that illustrate the
opposite point, but that you are not told
about. There are cases of eminent men
who had rather small skulls. ' But that
does not make good press.
* Witelson, Kigar, & Harvey, 1999.
' Ganis et al., 2004; Piazza & Dehaene, 2004.
* Deary et al., 2007.
5 Poltrock & Brown, 1984.
** Simonton, 2006.
r See Gould, 1981, pp. 92 ff. for examples.
But what are “comparable animals”? The
A and rg parameters differ across classes of
mammals. Therefore, it is more useful to
compare the relative encephalization indices
between two species, within a reference
class. First the encephalization ratios are
computed for all species within the class
of interest. Then one species is arbitrarily
assigned an encephalization value of one.
Other species in the class are compared to
that species, according to the equation
where j is the species of interest and i is the
index species for the reference group.
Within the nonhuman primates, reports
of innovation, tool use, and social reason¬
ing increase with relative brain volume. The
correlations are in the .5 range and above.7
7 Reader & Laland, 2002.
This suggests taking a closer look at our own
reference group, the great apes and, more
specifically, the hominids.
Using modern humans (. Homo sapiens )
as the index species [Rhuman = 0 f°r the
great apes, the chimpanzee has a relative
encephalization index of .3. Australopithicus,
the genus that preceded the genus Homo,
and that last lived about 2.3 million years
ago, had a relative encephalization index in
the .4 to .45 range. Our immediate evolu¬
tionary ancestors, Homo erectus, first appear¬
ing about 1.8 million years ago, had an
encephalization ratio of slightly less than .8.
The fossil record indicates rapid increases
in encephalization beginning about 500,000
years ago, with the first appearance of rep¬
resentatives of Homo sapiens. Our sometime
contemporary, but now extinct “cousins”
(i.e., descendants of a common ancestor,
but not ancestral to us), the Neanderthals,
had an encephalization ratio close to 1. (The
exact value is hard to determine because
i86
HUMAN INTELLIGENCE
we have only a few adequate fossils. Nean¬
derthal brains were somewhat larger than
modern human brains, but their bodies were
also larger.) Encephalization appears to have
peaked within Homo sapiens about 20,000
years ago, and may have decreased slightly
since then.8
If you accept the idea that the progression
from throwing sticks and making stone tools
to throwing bombs and making plastic tools
represents an increase in intelligence, then
there is clear evidence that across species
increases in brain size parallel an increase in
intelligence. But does this relationship apply
to individuals within the same species, us?
7.4.2. Intelligence and Variations in Brain
Size in Humans
In the early nineteenth century there was
widespread interest in phrenology, a pseudo¬
science whose followers claimed that they
could diagnose intelligence and personal¬
ity from the size and shape of the skull.
They had little evidence to back up their
assertions. The first objective evidence for
a within-species brain size:intelligence rela¬
tionship was obtained when Galton, and
then his colleague Karl Pearson, found a cor¬
relation of .11 between Cambridge students'
skull sizes and their grades.9
Galton faced a handicap. His study par¬
ticipants were alive. I am not being entirely
facetious. In Galton' s time the only way to
measure cranial capacity, and by inference
brain size, in a living person was to make a
measure on the exterior of the skull. This
method of measurement confounds thick¬
ness of the skull and cranial capacity.
To overcome this handicap several
researchers, both in Galton 's time and since,
have estimated brain size by direct measure¬
ment of cranial capacity post mortem. This
line of research was so popular in the late
nineteenth and early twentieth centuries
that it led to a lively - and, to the modern
eye, quaint - technical dispute over the rela¬
tive merits of measuring brain size by filling
8 Geary, 2005, pp. 50-54.
9 Citation by Rushton, 1992.
the cranial cavity with bird seed, shotgun
pellets, or sand. An obvious shortcoming
of such studies is that the investigators sel¬
dom had measurements of how intelligent
the people whose crania they were exam¬
ining had been when those people were
alive. This did not stop the early researchers.
They simply assumed that certain groups,
usually Europeans as compared to non-
Europeans, but sometimes men as compared
to women, were smarter than others. The
investigators then pointed to intergroup dif¬
ferences in cranial capacity as proof of their
assumptions.
In 1981 the Harvard paleontologist
Stephen J. Gould wrote an elegant, amusing,
and scathing review of this line of research,
as part of a more general, and equally
scathing, analysis of virtually all aspects of
research on intelligence.10 He concluded
that there was no reliable evidence relating
brain size to intelligence. He also claimed
that attempts to show group differences in
brain size were motivated by racial preju¬
dice. Gould had previously achieved con¬
siderable public credibility as a commen¬
tator on science, so his views were widely
accepted in spite of negative reviews of his
work in the technical literature.11
It is impossible to know whether Gould's
imputations about investigators’ motives
were correct or not. Indeed, if the early
investigators' facts were right it does not
matter to science what their motives were.
There have been several careful studies
about the brain size-intelligence relation¬
ship. Gould had his facts wrong.
In 1992 J. Phillipe Rushton, a professor at
the University of Western Ontario, analyzed
records of the cranial capacity in American
service men and women, taken as part of
the physical examinations for military ser¬
vice. He found that the mean intracranial
volume in officers was greater than that in
enlisted men, even after considering body
size. It is tempting to say that what one
makes of this depends upon whether one
10 Gould, 1981, 1996.
11 For an exceptionally comprehensive review, see
Carroll, 1995.
INTELLIGENCE AND THE BRAIN
187
served as an officer or as an enlisted per¬
son. More seriously, though, there is little
doubt that the average IQ test score for offi¬
cers is higher than the average for enlisted
service men and women. In 1996 Rushton
and his colleague C. D. Ankney published a
review of similar studies that had been done
to that date. They concluded that there is a
correlation of .15 between estimates of cra¬
nial capacity based on skull measurements
and measured intelligence.12 Rushton also
found differences between men and women,
and between racial groups. These are dis¬
cussed in Chapter 11.
In 1991 Lee Willerman, of the Univer¬
sity of Texas (Austin], reported a reliable,
positive correlation between intelligence
and brain size measured using imaging
techniques.13 Willerman’s study was fol¬
lowed by others. In 2004 a meta-analytic
review concluded that the brain size-IQ
correlation is about .33. 14 As the correla¬
tion between external estimates of the size
of the crania and imaging estimates of the
volume of the brain is about .5, we would
expect the cranial estimate-IQ correlation
to be approximately .16, in agreement with
Rushton and Ankney’ s finding. More recent
studies have suggested that the relationship
is primarily driven by the density of gray
matter, indicating the importance of local
connections within each region of the brain.
However, there does appear to be a smaller,
but reliable, correlation with the density of
white matter.15
As far as I have been able to determine,
there is only one study in which investigators
were able to correlate post mortem mea¬
sures of brains with levels of intelligence.16
Such a study is important, because direct
measures of parts of the brain are more accu¬
rate than the estimates obtained through
imaging. The results were confirming in
some ways and confusing in others. There
was a .6 correlation between overall brain
volume and measures of verbal intelligence,
12 Rushton and Ankney, 1996.
13 Willerman et al., 1991.
14 McDaniel, 2005.
15 Luders et al., 2009.
16 Wittelson, Beresh, & Kigar, 2006.
in right-handers. This is on the high side of
the measures reported in the various meta¬
analyses. There was a much smaller relation
between spatial-visual reasoning and brain
volume in women, and essentially none in
men. These findings indicate that in addi¬
tion to size differences there may be quali¬
tative differences in the way that the brain
is organized in men and in women, and in
right-handers as compared to left-handers.
We explore this issue in the next section.
7.4.3. Structural Differences in Various
Regions of the Brain and Their Relation
to General Intelligence
While it would be naive to think that intel¬
ligence is located in any one place in the
brain, it is reasonable to believe that differ¬
ent parts of the brain participate to differ¬
ent degrees in producing intelligence. Brain
density can be measured, on a region-by¬
region basis, using modern imaging tech¬
niques. When this is done there is a shift
from measuring volume to measuring den¬
sity of either gray or white matter, as dis¬
cussed earlier. Definitive studies are diffi¬
cult to accomplish, because the expense of
imaging makes it difficult to obtain a large
sample. This reduces statistical power, lead¬
ing to a concern that small to moderate
correlations may not be discovered. Never¬
theless, some good studies have been done,
and some things have been learned. Unfor¬
tunately, the results do not present a clear
picture.
A wide-ranging review concluded that
there were substantial correlations between
measures of cortical density and intelligence,
in a large number of cortical areas.1"
Three studies have been conducted on
large samples, chosen to be representative
of the populations of the US (two stud¬
ies] and Spain (one study).18 These studies
involved correlations using a general mea¬
sure of intelligence, either g or a compos¬
ite score, such as the full-scale intelligence
17 Luders et al., 2009.
18 Colom et al., 2009; Haier et al., 2009; Karama et al.,
2009.
i88
HUMAN INTELLIGENCE
quotient (FSIQ) of the WAIS. Whether
or not the results were consistent depends
on how closely you look. All the stud¬
ies found that the density: IQ correlations
were highest for areas within the frontal
and parietal regions, and in the cingulate
gyrus and the limbic system. However, they
disagreed about precisely where the activ¬
ity was within these broad brain areas. To
illustrate, in one of the studies there were
reliable correlations between density of gray
matter and a g score in over forty different
local regions of the brain. In another, using a
different test battery that had a similar factor
structure, there were reliable correlations in
fifteen different areas. Reliability hardly cap¬
tures the result; the probability levels for the
findings are in the p < .005 region. How¬
ever, there were only six areas of overlap,
in which the same region was implicated in
both studies.19
This is consistent with an earlier review,
in which Richard Haier, at the University
of California, Irvine, and his colleague Rex
Jung, at the University of Arizona, deter¬
mined the percentage of studies in which
reliable relations had been found between
the volume or density of a brain region and
a measure of intelligence.20 There was only
one area of the brain, in the left inferior pari¬
etal lobe,21 where 50% of the studies of brain
imaging had shown a relation between size
and intelligence. The densities of areas in
the anterior frontal cortex, other parts of the
parietal cortex, and regions in the cingulate
cortex were correlated with general intelli¬
gence scores in 30% of the studies. Most of
the areas showing effects were in the left
hemisphere, reinforcing the idea that the
sorts of information processing required for
conventional intelligence tests and reason¬
ing problems are carried out primarily by
the left side of the brain.
Why would such inconsistent results be
obtained? There are three possible explana¬
tions. The least interesting explanation, but
one that cannot be ruled out, is a statistical
19 Haier et al., 2009, Table 4
20 Jung & Haier, 2007. See especially their Figure 2.
21 Broadman area 40.
one. In this research correlations are calcu¬
lated between many densities of many dif¬
ferent brain areas and some form of IQ score.
In order to avoid reporting results that are
high “by chance/’ investigators adopt strin¬
gent statistical criteria. This increases the
chance that low to moderate correlations
will be overlooked. The second possibility
is that different people’s brains are actu¬
ally organized in somewhat different ways.
The differences would not have to be major
ones. There are changes in brain organiza¬
tion during childhood, and there is differ¬
ential organization across men and women.
There could be other small, but systematic
changes due to other demographic variables.
There is also a psychological possibility.
If a g measure is obtained using a battery-
type test, such as the WAIS, an individual’s
score is determined by a weighted compos¬
ite of scores on the subtest batteries. There¬
fore, two individuals can obtain the same
g or IQ score in two different ways - for
example, one person could have a high IQ
score on the WAIS by obtaining a high ver¬
bal IQ and moderate performance IQ, while
another person could obtain the same score
with a moderate verbal score and a high
performance IQ. More generally, when we
ask someone to demonstrate “intelligence”
they do not do so, because intelligence is
an abstraction. They execute a medley of
information-processing acts, using different
parts of the brain for the different acts. If a
g measure is obtained from a single marker,
such as a progressive matrix test, the ques¬
tions on the test will be fairly complex, and
will yield to different strategies. The same
principle applies as in the case of the battery-
type test; two people may obtain the same
score using different elementary processing
actions. When this is the case they will not
display the same pattern of brain activa¬
tion, even though they obtain equivalent
test scores.
7.4.4. Efficiency Counts
We have been discussing measures of struc¬
ture. An alternative way to look at the rela¬
tion between intelligence and the brain is
INTELLIGENCE AND THE BRAIN
189
to examine measures of overall neural effi¬
ciency, thus exploring the idea that in addi¬
tion to having bigger brains, intelligent peo¬
ple may have brains made of a better grade
of neurons.
The rudiments of this idea can be traced
back to Galton’s studies of reaction time. It
is also captured in modern studies of rela¬
tions between intelligence and the speed of
information processing during very simple
tasks, as discussed in Chapter 6. However,
the evidence connecting overt response
speed to neural measures is at best indi¬
rect. Studies that have used the new imaging
technologies to examine brain action dur¬
ing thinking are more informative, especially
since it was not clear in advance what they
should have found.
Suppose two people attack the same
problem. One of the problem solvers has
markedly higher intelligence test scores than
the other one. Would you expect the person
with the higher scores to show more or less
metabolic and neural activity than the per¬
son with the lower scores?
A case can be made for either answer.
A high metabolic rate might be a sign that
highly intelligent people have more mental
energy in a literal sense, for their metabolic
systems might provide more energy to their
brains. A low metabolic rate might be a sign
that highly intelligent people have more effi¬
cient neural systems, and therefore need less
energy per mental computation than less
intelligent people do.
In 1988 Richard Haier's University of
California, Irvine, group showed that the
“more efficient’' hypothesis is correct. They
obtained PET scans from eight univer¬
sity students as the students solved Raven
Advanced Progressive Matrices (RAPM)
problems. The correlations between meta¬
bolic rates and RAPM scores were in the
— .7 range, varying somewhat across areas
of the brain.22 As a result of the finding by
Haier’s group, the weight of the evidence
shifted toward the hypothesis that intelli¬
gence is associated with brain efficiency, not
energy generation.
22 Haier et al., 1988, Table 2.
Haier and colleagues' findings have been
replicated, using fMRI and other techniques.
A particularly interesting line of research has
been followed by Austrian researchers using
the very different EEG/ERP technology.23
They found that the amplitudes of neural
responses to verbal or spatial problems were
negatively correlated with measures of ver¬
bal or visual-spatial reasoning, respectively.
There was also an interesting male-female
difference. The relations between EEG
amplitude and test scores were strongest for
males attacking spatial-visual problems and
females attacking verbal problems, mirror¬
ing the differences between men and women
in verbal and visual intelligence.
A study from Carnegie-Mellon Univer¬
sity also bears on the issue of efficiency.
University students were given the reading
span test described in Chapter 6. On the
basis of their scores they were divided into
groups having high or low verbal working
memory. All groups then attempted sen¬
tence comprehension tasks, with sentences
that varied widely in their linguistic com¬
plexity (e.g., sentences with high or low fre¬
quency words, simple or complex syntax].
Their brains were imaged using fMRI during
the sentence comprehension task. Activa¬
tion in areas associated with linguistic anal¬
ysis decreased with working memory span,
and increased with linguistic complexity.24
Two further studies are worth noting,
because they illustrate the flexibility of the
brain. In addition to finding that people with
high verbal working memory spans showed
less overall activation, the Carnegie-Mellon
group also found that the high-span individ¬
uals showed a greater response to increas¬
ing linguistic complexity than did low-span
individuals. There was evidence of greater
coordination in activation of different areas
of the brain in the high-span than in the
low-span group.
The second study, by a different group,
utilized the “three back” task. In this task
a series of verbal or figural stimuli are
23 Neubauer, Fink, & Schrausser, 2001; Neubauer et al.,
2005.
24 Prat, Keller, & Just, 2007.
190
HUMAN INTELLIGENCE
presented. The participant is supposed to
respond if the current figure is identical to
one presented three items back in the series.
As an example, using letters, suppose that
the series was
ABCARQKKQ.
A “three back” response should be given
to the second A and the second Q. These
are called targets. The second K should not
be responded to, because it is “one back”
rather than “three back.” Such items are
called lures. Lures are frequently misiden-
tified as targets. It evidently requires an
effort to suppress responding, for £MRI scan¬
ning showed that when lures are presented
there is an increase in metabolism in the
frontal and parietal areas. This was inter¬
preted as an indication that an inappropri¬
ate response was being suppressed. Scores
on the Raven’s Matrix test were positively
correlated both with accuracy and with the
size of the metabolic response.25
These results are typical of other find¬
ings. Several research groups have reported
that when they contrast metabolic or elec¬
trical activity from people who, by other
means (intelligence tests or, often, measures
of working memory] have been shown to
have different levels of cognitive skills, the
more skilled tend to activate fewer brain
sites.26
What happens when the brain is
“on idle?” Haier’s UCI group asked this
question.27 They measured brain activa¬
tion while people were watching videos,
but not otherwise engaged. In this admit¬
tedly exploratory study, high Raven Matrix
scores were associated with greater inte¬
gration between the object-recognition and
linguistic-processing areas of the brain.
There is an interesting resemblance between
this result and Ackerman’s contention (dis¬
cussed in Chapter 5) that intelligent people
25 Gray, Chabris, & Braver, 2003.
26 Bomkessel, Fiebach, & Friederici, 2004; Caplan,
Waters, & Alpert, 2003; Neubauer et al., 2005; Lar¬
son et al., 1995.
27 Haier, White, & Alkire, 2003.
show an above-average intellectual engage¬
ment with tasks that are not directly related
to their work or study. Even when watch¬
ing videos from commercial television, the
intelligent brain does not completely disen¬
gage.
Both the structural and process technolo¬
gies present the same basic message. Intel¬
ligence is associated with larger brains, and
with more efficient brains. However, there
is no single hot spot in the brain, associ¬
ated with all aspects of cognition. The brain
provides a tool kit for intelligent action.
An intelligent person has somewhat higher
quality tools, and organizes them more effi¬
ciently, than an unintelligent person.
7.5. The Brain and Specific Cognitive
Functions
Accepting the idea that the brain is a tool
kit, we now look at some of the special¬
ized tools. We first look at the relation
between working memory and brain func¬
tions, on the grounds that working memory
is tightly enmeshed with general reason¬
ing, and hence our most important sin¬
gle general processing capacity. We then
look at the brain structures underlying other
information-processing capacities that have
been associated with various aspects of
intelligence.
7.5.1. The Brain and Working Memory:
Evidence and the P-FTT Model
Figure 7.3 summarizes the findings of
over two hundred studies in which brain
metabolism was measured as people did var¬
ious activities related to attention and rela¬
tively short-term memory, including oper¬
ations on information while it is being held
in memory.28 The reviewers summarized the
study as showing that working memory tasks
showed heavy involvement of the frontal
and parietal cortices, and the anterior cingu¬
late gyrus. Lateralization depends upon the
28 Cabeza & Nyberg, 2004. See also Smith & Jonides,
1999
INTELLIGENCE AND THE BRAIN
i9i
Figure 7.3. A sketch of areas of the brain that show activity during a variety of tasks involving working
memory. The upper figures show a lateral view of the cortex; the lower figures show a medial view.
From Cabeza & Nyberg, 2000, Figure 3. Reprinted with the permission of Massachusetts Institute of
Technology Press Journals.
nature of the stimuli. In general, problems
involving spatial and figural stimuli produce
greater activation in the right than in the
left hemisphere, while the reverse pattern is
seen for problems involving language.
Figure 7.3 shows what parts of the brain
are involved in working memory tasks in
general. But to what extent are these areas
related to individual differences in intelli¬
gence, and especially to either (depending
on your theoretical predilections) g or Gf?
One approach to this question is to take
tasks that are known to have high g load¬
ings in the intelligence literature, such as
progressive matrix tests, and to determine
what areas of the brain are active when
people do these tasks. Another strategy is
to determine what sort of brain injuries
result in selective loss of the ability to deal
with Gf-type problems, as opposed to Gc-
type problems, where the solution depends
largely on retrieving previously acquired
information.
Both approaches lead to essentially
the same answer.29 Jung and Haier have
29 Many studies could be cited. Some representative
ones are Duncan, 1996, and Duncan, Burgess, &
Emlie, 1995, for brain injury; Duncan et al., 2000,
produced the Parieto- Frontal Integration The¬
ory (P-FIT) theoretical model that does
a good job of encapsulating our present
knowledge.50 They propose that our abil¬
ity to do the sort of thinking captured by
measures of Gf and working memory is sup¬
ported by a system of brain regions involv¬
ing the dorso-lateral frontal cortex, the
parietal lobe, and the anterior cingulate
gyrus. Each of these regions performs some
of the functions needed for abstract reason¬
ing and problem solving; no one of them is
sufficient alone.
Jung and Haiers idea of the role of
the frontal cortex is consistent with other
observations, both of imaging results and of
the sort of scattered thinking displayed by
Phineas Gage, and by many other patients
with frontal lobe damage. In general, the
frontal lobes seem to be necessary to keep
a person on-taskT1 In one study that nicely
and Haier, White, & Alkire, 2003, for differential
activation of brain areas during imaging; and Colom,
Jung, & Haier, 2006, for a discussion of the relation
between neural density and g loadings.
30 Jung & Haier, 2007.
31 See Duncan et al., 1996, for a good discussion of
this point. Shallice (2004), provides an independent
review of frontal lobe functioning that is generally
consistent with Jung and Haier s view.
192
HUMAN INTELLIGENCE
illustrates the effect, it was shown that chil¬
dren who did not have a history of attention
disorder prior to injury to the frontal and
especially prefrontal-cortex displayed symp¬
toms of attention deficit disorder after the
injury.32 One of the intriguing findings on
this topic is that one area of the frontal lobe
appears to be responsible for orchestrating
thinking about things and abstract ideas,
while another region orchestrates thinking
about socially relevant topics.33
Jung and Haier argue that the parietal
cortex is responsible for integrating infor¬
mation from various sensory modalities.
This would be consistent with the parietal
cortex’s established role in controlling the
deployment of attention externally, to par¬
ticular regions of and objects in the sensory
fields.34 The role of the parietal cortex in
providing temporary storage areas for infor¬
mation also appears to be well established.
This seems to be an area where the lateral¬
ization is especially well marked; linguistic
information, the phonetic loop in Baddeley's
model, resides in the left (in most of us),
while spatial and object information is held
on the right.33
Jung and Haier propose that the ante¬
rior cingulate gyrus acts as a response selec¬
tion device. It is responsible for directing
decisions, albeit with substantial regulatory
input from the frontal lobes. The ante¬
rior cingulate gyrus also seems to weigh
the likely consequences of taking an action.
Clancy Blair, a developmental psycholo¬
gist at Pennsylvania State University, has
pointed out that recent research, which
it would take us too far afield to exam¬
ine, has shown that emotional evaluation
of outcomes plays an important part in
response selection, even in situations where
one would expect rational decision making
to be the norm.36 Combining Blair and Jung
and Haier’s views, an important part of the
frontal cortex-cingulate gyrus interaction
32 Max et al., 2005.
33 Beer, Shimamura, & Knight, 2004.
34 Posner et al., 2006.
35 Cabeza & Nyberg, 2000. For a good illustrative
study, see Smith & Jonides, 1999.
36 Blair, 2006.
may be modulation of the emotional and
nonemotional aspects of decision making.
In summary, there is clear evidence that
the working memory system, which we
know is central to reasoning and general
intelligence, is supported by a brain sys¬
tem involving regions of the frontal lobe,
the parietal lobe, and the anterior cingulate
cortex. I do not want to give the impression
that these are the only areas involved, or that
all the details of the involvement have been
worked out. They have not, but the outline
is clear.
Having disposed of g, we now look
at brain correlates of the three secondary
dimensions of the g-VPR model: verbal intel¬
ligence, perception, and spatial rotation.
The discussions of perceptual and spatial-
rotation dimensions will be collapsed into
one, with an emphasis on the R dimension.
This is because the imagination of move¬
ment in space is central to spatial reasoning,
which seems to me more intellectual than
detecting small differences in visual stimuli,
and because the ability to visualize move¬
ment appears to be related to mathematical
and numerical reasoning.
7.5.2. Structures Associated
with Verbal Intelligence
Where is language in the brain? The answer
to this question can be brief, not because
so little is known but because so much is
known that reference can be made to general
textbooks and reviews. There is also what is
known as the “issue of granularity.” When
we talk about a person having high verbal
intelligence, we mean that the individual
displays a high level of general competency
with language. This includes a large vocabu¬
lary, and the ability to follow complex argu¬
ments. We do make a distinction between
comprehension and expression, acknowl¬
edging that some people who compre¬
hend well do not always express themselves
well.
At this level of detail the neuroscientific
basis of normal language comprehension has
been known since late in the nineteenth cen¬
tury. There are major centers for language
INTELLIGENCE AND THE BRAIN
x93
Figure 7.4. Broca’s and Wernicke’s regions on
the left side of the brain. The display is for a
right-handed individual. Sketch by the author.
comprehension in the left posterior frontal
lobe (Broca’s area] and in the left poste¬
rior temporal lobe (Wernicke's area]. These
are shown in Figure 7.4. These regions were
identified on the basis of clinical studies.
Patients with lesions in Broca’s area under¬
stand language but are unable to express
it. Patients with lesions in Wernicke’s area
can speak, but their sentences are often
incoherent.
Imaging studies have confirmed the
nineteenth- and twentieth -century neuro¬
psychological findings. Figure 7.5 shows a
summary of regions where imaging studies
have identified brain activation associated
with the perception and naming of words.
While there is some involvement of the right
hemisphere for listening, without a spoken
response, the majority of the active centers
are clearly in the left posterior frontal and
temporal regions.
A similar picture is obtained in stud¬
ies showing relations between verbal com¬
prehension test scores and neural density.37
One project (covering two studies, so self¬
replication was accomplished] obtained cor¬
relations greater than .7 between the density
of gray matter in the posterior left temporal
region and participants’ scores on the vocab¬
ulary and information subtests of the WAIS,
the two tests most associated with verbal
comprehension.38 Another has reported that
adolescents with high scores on the vocabu¬
lary subtest of the WAIS show higher den¬
sities of gray matter in, somewhat surpris¬
ingly, the parietal lobe.39 Two other reports
37 Colom et al., 2009. Gc tasks tend to stress language
ability (Carroll, 1993].
38 Colom et al., 2006, Tables 2 and 4.
39 Lee et al., 2007.
have indicated that reduced neural density
in the temporal and frontal lobe regions
associated with language identifies young
children at risk for developing reading
difficulties.40
If we move to a more complex task,
sentence comprehension, we find a simi¬
lar picture, with one addition. There begins
to be more activation of the left frontal
regions, those regions that are also involved
in working memory tasks involving words or
sentences.41 This is hardly surprising. One
of the more important findings, though, is
that the more complex the sentence com¬
prehension task the greater the activation
level in all areas associated with language
comprehension.42 These findings are now so
well established that research has moved for¬
ward to attempts to locate particular spe¬
cialized functions - for example, process¬
ing a verb - rather than repeating studies
looking for the general location of language.
At this point students of individual differ¬
ences are apt to lose interest, because of the
need to maintain Brunswikian symmetry by
trying to correlate concepts that are at the
same level of granularity. Verbal intelligence
is a broader concept than comprehending
verbs.
7.5.3. Structures Associated with
Perceptual and Rotational Skills
Verbal and Perceptual and Rotational
skills (more properly, imagery and spatial-
reasoning functions] are produced by differ¬
ent parts of the brain. To explain this we
must consider how the brain conducts visual
analysis.
Visual stimuli are initially analyzed in the
optic tract and connected subcortical struc¬
tures, prior to arriving in the occipital lobe,
at the back of the brain. Analyses in the
occipital lobes construct three-dimensional
representations of an object from the two-
dimensional pattern on the retina. The
40 Deutsch et al., 2005; Hoeft et al., 2007.
41 Caplan, Waters & Albert, 2003; Colom et al., 2009;
Cooke et al., 2006.
42 Prat, Keller, & Just, 2007.
i94
HUMAN INTELLIGENCE
Figure 7.5. Areas that have shown activation during selected language comprehension tasks. Note the
distinction between visual and auditory input, and between reception and reception accompanied by
a verbal response. The upper figures show a lateral view of the cortex; the lower figures show a
medial view. From Cabeza & Nyberg, 2000, Figure 6. Reprinted with the permission of Massachusetts
Institute of Technology Press Journals.
(interpreted) visual image is then analyzed
by two neural pathways. One goes from the
occipital lobe along a dorsal route through
the parietal lobe; the other moves ventrally
into the temporal lobe. (See Figure 7.6). The
ventral pathway is largely concerned with
identifying what the stimulus is, including
identification of its attributes (e.g.; color).
The dorsal pathway is concerned with where
the stimulus is, and whether or not it is mov¬
ing. Individual differences in visual-spatial
reasoning, Gv in the three-stratum model
Figure 7.6. The dorsal and ventral visual
streams. The ventral stream, through the
temporal lobe, is primarily concerned with the
identification of stimuli and their attributes. The
dorsal stream carries information about location,
movement, and configuration. Sketch by the
author.
and the PR dimensions in the g-VPR model,
are related to gray matter cortical density
along both pathways.43
Imagery and visual reasoning utilize brain
structures that are close to, although not
exactly identical to, the structures that sup¬
port vision. Thus, as a rough approximation,
we can think of the brain's visual analysis
system as being driven either by the visual
sensory system or by some “executive con¬
troller” in the brain itself.44 In terms of the
tool kit analogy, different parts of the brain
can be used to attack a task, depending upon
whether a verbal or a spatial-visual strat¬
egy is chosen. Panel 7.5 contains an example
illustrating this point.
Male and female brains deal with visual-
spatial tasks in somewhat different ways.
This is interesting because there are marked
male-female differences in visual-spatial rea¬
soning, especially if the task involves visu¬
alizing motion. There is also a substan¬
tial amount of evidence indicating that the
level of androgens influences visual-spatial
reasoning. Prenatal influences seem to be
43 Colom et al., 2009, Figure 4. Note the Gv correlates
indicated in this figure.
44 Ganis et al., 2004; Kosslyn, 1994.
INTELLIGENCE AND THE BRAIN
*95
Panel 7.5. The Sentence Verification
Paradigm and its Neural Correlates
The sentence verification paradigm has
been widely used to study verbal com¬
prehension. The participant first reads a
sentence describing a picture and then
sees the picture. The task is to deter¬
mine whether or not the sentence accu¬
rately describes the picture. Quite simple
pictures are used. A favorite is the “plus
above star” arrangement, where the pic¬
ture is either
The simplest descriptions of these pic¬
tures are “star above plus” and “plus above
star.” More complicated sentences can be
used, for example, “plus not below star.”
In either case the task can be regarded
as a prototypical linguistic act, where the
reader must coordinate verbal and per¬
ceptual information.
There are two strategies for sentence
verification. In the visual strategy the par¬
ticipant reads the sentence and creates an
image of the anticipated picture. When
the picture is shown the participant
decides if it matches the image. In the
verbal strategy the participant first reads
and memorizes the sentence. When the
picture is shown the observer describes
it, covertly, and decides whether or not
the verbal description has the same mean¬
ing as the memorized sentence. College
students can use either strategy.5" Marcel
Just’s group at Carnegie-Mellon univer¬
sity took fMRI images of the same peo¬
ple when they were instructed to use
either the verbal or visual strategies, thus
relying on either verbal or visual work¬
ing memory. * The centers of activation
were in either the left posterior frontal
region (verbal strategy) or the right pari¬
etal region (visual strategy), depending
on the strategy used. Figure 7.7 illustrates
the result.
This experiment shows that you can¬
not say that the brain controls this or that
piece of cognitive behavior. It depends
upon how the behavior is achieved.
* MacLeod, Hunt, & Mathews, 1978; Mathews,
Hunt, & MacLeod, 1980.
Reichle, Carpenter, & Just, 2000.
especially important. To show how general
this is, the phenomenon can be produced
in rats and mice. The situation seems to be
quite complicated, but there is little doubt
that individual differences in brain function¬
ing do affect visual-spatial reasoning.
Orienting ability , the ability to develop a
mental map of an area and to locate one¬
self in it, is worth special discussion. Ori¬
enting ability is certainly part of intelligence
in the conceptual sense, but it is assessed
only indirectly by conventional intelligence
tests, due to the limitations of the conven¬
tional “Drop in from the Sky” testing format.
This is one of the interesting cases where the
argument that learning must influence the
brain can be supported by evidence. London
taxicab drivers show enlarged hippocampi,
compared to appropriate control subjects.45
In a more controlled setting, hippocampal
volume is positively associated with how
well people learn to navigate a computer¬
generated virtual environment.46
7.5.4. Long-Term Memory
Working memory is the workbench used to
keep track of things related to the problem
that is before us. Long-term memory refers
to the ability to acquire some information,
put it aside for awhile (which may be any¬
where from a few minutes to a year), and
then recall it when required. What brain
structures are involved, and how do they
45 Maguire et al., 2003.
46 Moffat et al., 2007.
196
HUMAN INTELLIGENCE
relate to individual differences in long-term
memory? It turns out that the answers to
these questions depend upon the sort of
memory we are talking about. We have to
distinguish between three different types of
memory, and two different recall mecha¬
nisms.
Episodic memory refers to memory for
specific events that have occurred in one’s
life - for instance, what you had for breakfast
yesterday morning. Semantic memory refers
to memories of how the world works; know¬
ing that "hens lay eggs” is a piece of seman¬
tic memory. Procedural memory refers to
knowledge of how to do things. Riding
a bicycle is an often-used example. Logi¬
cally, one might think that episodic mem¬
ory would be central, and that semantic and
procedural memories would derive from it.
This is not the case. It is possible to acquire
semantic and procedural information with¬
out storing an episodic record of how that
information was recalled. The issue is not
one of forgetting, which is normal. (Do
you remember when you learned that hens
lay eggs?) In some circumstances semantic
and procedural information can be acquired
even though no episodic record is ever laid
down. To explain this we have to look at
recall.
Recall can be explicit (declarative) or
implicit. Recall is explicit if you are aware
of the act of recall. If I ask you what you had
for breakfast, you will be aware of recalling
the answer. Whether you recall accurately is
another matter.
Implicit recall is shown by a demon¬
stration that behavior has been altered by
an experience, even though that experi¬
ence is not available for explicit recall.
Implicit recall is dramatically illustrated by
patients who have experienced electrocon¬
vulsive shocks to the brain, a form of ther¬
apy that actually has a history of benefits
in some psychoses. The patient will typ¬
ically not remember the experience but
does show signs of nervousness if he or
she reenters the room where the therapy
took place. However, implicit recall is not
solely associated with pathologies. It can be
demonstrated in an undergraduate psychol¬
ogy laboratory, using quite innocuous pro¬
cedures.47
Studies associating individual differences
in memory with intelligence generally deal
with explicit recall of episodic and semantic
information. Many episodic memory stud¬
ies deal with very short time intervals, often
only a few seconds, as in the digit span test,
which is part of many intelligence test bat¬
teries. Explicit recall of semantic memory
is also tested, often under the title “test of
general knowledge” or "vocabulary.” It could
be argued that semantic memory is a proto¬
typical example of a crystallized intelligence
(Gc) skill.
We have already seen, in the case of H. M.
(panel 7.4) that someone who has lost his
or her hippocampus will literally not recall
a meeting that took place fifteen minutes
earlier. H. M. had lost episodic memory.
Similar dramatic losses of memory have
been related to damage to the limbic system
that is associated with alcoholism.48 In these
cases the frontal lobe may also show signs
of damage. Dramatic losses of episodic and
then semantic memory occur in Alzheimer’s
disease, a variety of senile dementia that can
occur as early as fifty, affects 3-4% of peo¬
ple between sixty- five and seventy-four, and
approximately half of all Americans over
the age of eighty-five. Alzheimer’s disease
is associated with widespread reduction in
brain volume, including both the frontal
lobes and the hippocampal formation.
Are variations in brain structure also asso¬
ciated with individual differences in mem¬
ory within the normal range? This is a dif¬
ficult question to answer, because we have
to distinguish among the various types of
memories and how they are developed.
47 The technique used is to present people with very
long lists of words, more than they can recall. Sub¬
sequently they are asked to perform some task that
tests their readiness to use a word on the list, or some
other word. There will be a tendency to use the
word on the list, even though people cannot recall
seeing it on the list. See Jacoby, Toth, & Yonelinas,
J993-
48 Sullivan & Marsh, 2003.
INTELLIGENCE AND THE BRAIN
i97
Figure 7.7. Regions of the brain activated within a single
individual when that person was instructed to solve a sentence
comprehension problem using either a verbal or a visual imagery
strategy. The task is described in panel 7.5. From Reichle et al.,
2000, Figure 3, with permission from Elsevier.
Consider how memories come to be.
First, you have an experience. An internal
representation of that experience will be
created. This is a constructed representa¬
tion of what is going on, an act of prob¬
lem solving that takes place largely in the
frontal-parietal- anterior cingulate circuit.
The information in the coded representa¬
tion is then transferred to long-term mem¬
ory, along with its connections to informa¬
tion already in memory. This function is
carried out by the hippocampus and other
structures in the medial temporal cortex.
The hippocampus is essential for the storage
198
HUMAN INTELLIGENCE
process, but, with the exception of maps
of spaces, storage apparently takes place all
over the cortex.49
Given this two-stage process, it is not sur¬
prising to find that individual differences
in learning and memory have been asso¬
ciated with measures of volume in both
the frontal lobes and the hippocampus and
in related structures. The studies are small
(in part because they are expensive], so
there has been a tendency to focus on
extreme groups, but the picture is con¬
sistent. Both the frontal-parietal-cingulate
gyrus and the medial temporal cortex sys¬
tems are required - one to decide what is
to be remembered and one to carry out the
storage process.*0 As is the case for virtually
all findings in the rapidly developing field of
neuroscience, all the details are not filled in.
One thing is quite clear. Memories do
not reside in individual neurons; they reside
in patterns of neurons. Accordingly, it has
been suggested that intelligence depends
upon the brain's ability to establish new con¬
nections between neurons, as well as any
structural differences in the regions involved
in memorization. What is the evidence for
this?
7.6. Neural Plasticity
Human intelligence depends upon the abil¬
ity to learn. When we learn something we
change the brain, both in the sense that new
memories are stored in the brain and in the
sense that as we learn to do things (including
learning how to solve problems], the brain
is reorganized. The case has been made that
differences in the ability to reorganize the
brain in the face of experience, individual
differences in neural plasticity, may be an
important aspect of intelligence. Computer
simulations have shown that things could
work this way.*1 But do they?
49 Eichenbaum, 2004.
50 For typical work, see Habib, McIntosh, & Tulving,
2000; Rosen, 2003; Sullivan & Marsh, 2003; Tulving
et al., 1999.
51 See Garlick, 2002, for an example of this sort of
argument.
The effects of increasing intelligence
upon brain mechanism are mirrored by the
effects of increasing practice; the more prac¬
tice on a task, the lower the measured
metabolic rate. In another set of PET stud¬
ies by the UC Irvine group, Haier and his
colleagues had students learn to play the
computer game TETRIS, a highly demand¬
ing visual-spatial task. There was a marked
decrement in metabolism as the task was
learned. In addition, the greatest decrements
were observed in people with the highest
intelligence test scores. Similar findings have
been observed with quite different tasks.*2
Corroborating evidence comes from the
study of infant intelligence. There have been
many attempts to identify characteristics of
infant behavior that could predict later intel¬
ligence. One of the most successful proce¬
dures is a technique known as habituation.
In a habituation study an infant is shown a
picture, and then shown the now-familiar
picture along with a novel picture. Habitua¬
tion is measured by the extent to which the
infant looks at the novel picture. The idea
is that this tests the ability to form a mem¬
ory of the familiar picture, that is, to reor¬
ganize neural patterns to reflect experience.
Habituation measures taken in the first year
of life have a correlation of .36 (corrected
for unreliability, .53] with IQ scores taken
at age 21. 53
Studies like these suggest that there is a
relation between intelligence and the abil¬
ity to organize neural patterns over a brief
period of time. However, the neural reor¬
ganization is inferred rather than observed.
An important study using the diffusion ten¬
sor imaging technique provided direct evi¬
dence, over a much longer period of time.
During the early years of life the cerebral
cortex first thickens, then thins. The process
is not completed until late adolescence. It is
believed to be associated with selective re¬
arrangement of cortical neurons. The trajec¬
tory of development is much more sharply
52 Haier et al., 1992. See also the conceptual replica¬
tion, using a verbal memory task, by Habib, McIn¬
tosh, & Tulving, 2000.
53 Fagan, Holland, & Wheeler, 2007.
INTELLIGENCE AND THE BRAIN
m
defined in individuals with high IQ scores
(>120) than in individuals with lower
scores.54 This could be due to individual dif¬
ferences in the ability to reorganize the brain
to incorporate new information, or it could
be due to programmed (i.e., genetic) differ¬
ences in the developmental progression of
the neural system.
Many studies have shown that high intel¬
ligence test scores are correlated with faster
learning, in contexts ranging from labora¬
tory tasks through academic studies and on
to workplace apprenticeships. Any learn¬
ing has to be associated with some sort of
reorganization in the brain. Individual dif¬
ferences in neural plasticity could produce
such reorganization, but so could individual
differences in other brain mechanisms. For
instance, it could be that superior learning is
mediated by a superior ability to focus atten¬
tion, something that the intelligent clearly
have. Why intelligence is associated with
rapid learning remains in question. There
may not be any one answer.
7.7. What Do We Learn from Studies
of the Brain and Intelligence?
The brief answer to this question is "Lots,
and there is more to come.” A longer answer
is more thoughtful, and a bit more reserved.
Intelligence is associated with multiple
brain systems. General reasoning is sup¬
ported largely by circuits between the dor¬
solateral frontal cortex, the parietal cortex,
and the anterior cingulate cortex; verbal rea¬
soning by structures in the frontal and tem¬
poral lobes, largely but not exclusively in
the left hemisphere; spatial-visual reason¬
ing (including rotation) by the occipital and
parietal lobes; orientation by the hippocam¬
pus; and episodic memory by the general
reasoning system, to decide what to remem¬
ber, and by the hippocampus and related
structures that carry out the process of writ¬
ing information into storage locations all
over the cortex.
54 Shaw et al., 2006.
Frontal-parietal -
cingulate system
Posterior-frontal-
temporal system
Figure 7.8. The relation between brain systems
and broad (second-stratum) psychometric
abilities. The boxes above the dotted line show
relations between brain systems and the broad
abilities identified in the VPR model of
intelligence. The boxes below the dotted line
show the relations between brain systems and
cognitive skills that show strong individual
differences, but that are not typically evaluated
in psychometric studies due to limitations in the
conventional testing paradigm.
Three of these systems map closely
onto the dimensions identified in Johnson
and Bouchard’s VPR model of psychome¬
tric studies of intelligence.55 The relation¬
ship is shown, in block diagram, in the
region above the dotted line in Figure 7.8.
The frontal-parietal-cingulate gyrus sys¬
tem supports general reasoning, with some
hemispheric specialization for the modal¬
ity involved, typically verbal versus spatial-
visual reasoning. The system connecting
the occipital and parietal cortex supports
spatial-visual reasoning, while systems in the
posterior frontal and temporal cortex are
essential for verbal reasoning.
The systems approach is consistent with
several findings showing that the densities of
both gray matter and white matter correlate
with intelligence test scores. This includes
55 Johnson & Bouchard, 2005a.
General reasoning
Verbal reasoning
Perceptual -
rotational ability
Occipital -parietal
system
Individual differences in
memorization
Frontal-hippocampal
system
Parietal-hippocampal
system
Individual differences in
orientation
200
HUMAN INTELLIGENCE
both general scores, such as the Full Scale IQ
on the WAIS or the g composite on battery-
type intelligence tests, and scores on special¬
ized tests of verbal and nonverbal reasoning.
The density of gray matter probably reflects
the computing capacity within various spe¬
cialized computation centers in the brain,
while the white matter reflects the quality
of the connections between them.
The part of the diagram below the dot¬
ted line in Figure 7.8 shows relationships
between identifiable brain systems and func¬
tions that are clearly cognitive but are out¬
side the VPR model, even though there are
strong individual differences in them. The
deficit is not unique to the VPR model; it
is shared by all models of intelligence that
are restricted to an analysis of psychometric
data. Learning and orientation are certainly
part of the conceptual meaning of intelli¬
gence. Their evaluation has been excluded
from conventional testing simply because
the necessary behavioral evaluations do not
fit into the testing paradigm. This is a logis¬
tical rather than a scientific consideration,
and ought not constrain our thinking about
intelligence.
7.7.1. Intelligence Emerges from the
Interaction between Brain Systems
Terms like “general reasoning capacity” and
“verbal reasoning” refer to broad dimensions
of intelligence, Carroll’s56 second-stratum
abilities. These, by definition, can be applied
in many situations.
The broad dimensions can be broken
down further. This is essentially what stud¬
ies of individual differences in information
processing do. Figure 7.9 illustrates this, by
showing a possible further breakdown of
general reasoning ability. This diagram is
not to be understood as a seriously pro¬
posed model, although it has some support.
What it is intended to show is that con¬
cepts such as g and working memory refer to
properties that emerge from the interaction
between components that can be defined
56 Carroll, 1993.
at the information-processing level, and are
supported by separate neural mechanisms.
Similar diagrams could be drawn for ver¬
bal reasoning or for spatial-visual reasoning.
Because broad abilities emerge from the
interaction between components, attempts
to connect broad intellectual abilities with
brain-level and information-processing con¬
cepts should also deal with broad concepts,
such as the neural circuits supporting the
working memory and attention complex as
a whole. Attempts to relate a widespread
trait, such as g, to a component of the brain’s
functional networks, such as looking for the
neural locus of intelligence in some small
part of the frontal cortex, are not likely to
work out very well. It is probably a good idea
to remain at the appropriate system level, as
shown in Figure 7.8, rather than trying to
break the system into additive components.
This leaves us with the problem of
explaining the nature of the general intel¬
ligence factor, g. Arthur Jensen has said that
g is “a source of variance in performance
associated with individual differences in the
speed or efficiency of the neural processes
that affect the kinds of behavior called men¬
tal abilities.”57 This leaves open the ques¬
tions of “what neural processes?” and “what
behaviors?” for different cognitive actions
are supported by different brain systems.
Behaviorally, g is virtually synonymous
with general reasoning ability, which in turn
is synonymous with individual differences in
working memory. By this argument the seat
of g is in the frontal-parietal-cingulate cor¬
tex system. However, this raises some prob¬
lems. Vocabulary tests are highly g loaded,
but tests involving syntactical and seman¬
tic analyses of single words do not activate
the entire frontal-parietal-cingulate system,
and do activate areas outside of this system,
notably in the temporal lobe.58 The source of
g seems to jump around as the task changes.
Why?
57 Jensen, 1998, p. 74.
58 See Figure 7.5. For a typical study, see Friederici,
Opitz, & von Croman, 2000.
INTELLIGENCE AND THE BRAIN
201
Figure 7.9. Hypothetical relationships between brain systems,
narrowly defined information-processing functions, working
memory, and general reasoning ability (g) . This diagram is not
proposed as a model, but rather to show the complexity of the
issue, and the need to deal with broadly defined abilities, such as g,
as emerging from a system of interacting components, rather than
being a thing in itself.
7.7.2. What More Do We Need to Know
(and What Has Not Been Shonm)?
Investigations of the neural basis of intel¬
ligence are often more compelling than
behavioral studies. The chance to look at the
brains of intelligent and not-so-intelligent
people seems to many to be far more excit¬
ing than poring through analyses of cor¬
relation matrices. There certainly is more
to the neuroscience approach than just an
increase in excitement. The imaging and
enhanced electrophysiological technologies
available today provide researchers on intel¬
ligence with a new source of data. They can
rightly behave like nineteenth-century gold
miners, who rushed from California to Aus¬
tralia to the Klondike in an effort to strike
it rich in new fields. While a great many
nuggets of information have been mined,
there is no reason to believe that the neu¬
roscience gold field is about to pan out. It
does have its limitations.
Clearly overall brain size is correlated
with intelligence test scores. The correla¬
tion appears to be in the .3 range, which
is not everything, but is not to be dismissed.
I doubt that there will be any great effort
to develop this finding any more, because
of the findings on the differential impor¬
tance of various regions of the brain. These
findings, and the multivariate nature of
intelligence itself, make it clear that looking
at something as gross as the relation between
overall brain size and an omnibus intelli¬
gence score, such as the IQs derived from
batteries of subtests, as in the WAIS, has
gone about as far as it can go.
This chapter opened with an analogy
between the way the brain produces thought
and the way an orchestra produces music.
The cognitive orchestra contains several
functional players: working and long-term
memory, verbal reasoning, visual-spatial rea¬
soning, orientation ability, and numerous
subdivisions within each of these functions.
We now have a reasonably good idea where
each of these players sits in the brain,
although there are certainly details to fill in.
We also know that intelligence is associated
with slight structural changes and a marked
shift in the efficiency with which the brain
utilizes the players.
What we do not know is how the dif¬
ferent players in the brain work. Most cru¬
cially, we do not know how the conductor
works. It is certainly an advance in knowl¬
edge to move from saying “intelligence is
associated with big brains” to talking about
the locus of reasoning and executive control
202
HUMAN INTELLIGENCE
functions in the frontal, parietal, and cin¬
gulate gyrus. Nevertheless, such statements
are maps of where an activity is, not expla¬
nations of what it is. It is one thing to say
that the conductor is the person standing on
the podium at the front of the orchestra. It
is another to say that the conductor main¬
tains the tempo and signals emphasis for the
various orchestral sections. It is still another
thing to say how the conductor maintains
tempo, and so forth. We cannot stop by say¬
ing there is a conductor in, say, the system
specified by the P-FIT model. That is like
explaining the conductor by pointing to the
podium. We have to explain how the var¬
ious pieces of brain anatomy achieve their
functions. This remains a mystery.
The second mystery has to do with the
flexibility of the brain. To what extent is the
brain's development under the control of a
genetic program? How, and how much of
this development can be altered by expe¬
rience? Suppose we grant, for the minute,
that the unusual size of Einstein’s parietal
cortex was associated with his undoubtedly
superior mathematical talent. Was Einstein
able to do mathematics because his pari¬
etal cortex was large? Or did Einstein’s pari¬
etal cortex become large because he spent
a lifetime thinking mathematical thoughts?
On a more prosaic level, we know that
experienced London taxicab drivers have
large hippocampi and a superior sense of
orientation along London’s streets. Is this
because their hippocampi enlarged as they
acquired experience with London, or is it
because novice taxicab drivers with smaller
hippocampi could not develop the necessary
cognitive map, and had to find new employ¬
ment? We do not know.
Humans are provided, at conception,
with the potential for a brain. The poten¬
tial varies among individuals. Both in utero
and throughout their life individuals are
exposed to physical and social environments
that affect the development of the brain. In
the next two chapters we consider how that
development takes place.
CHAPTER 8
The Genetic Basis of Intelligence
Darwinian man, although well behaved
At best is a monkey shaved.
W. S. Gilbert
8.1. Introduction
The epigraph, taken from the 1909 Gilbert
and Sullivan operetta Princess Ida , almost
gets the scientific facts right. In the interest
of scientific accuracy the lines should read
Darwinian man, though well behaved
Is only an ape who's shaved.
for there are some key genetic differences
between apes and monkeys. We and our
other ape cousins have lost those ele¬
gant monkey tails, developed a different
limb structure, and have bigger brains. The
ape-monkey genetic changes took place
twenty million years ago; five millions years
ago more genetic changes produced the
hominids. Somewhere around 100,000 years
ago our own species, Homo sapiens, a very
big-brained species with a limb structure
that is fine for walking but terrible for
climbing trees, began to spread from our
place of origin in Africa, and by somewhat
less than 15,000 years ago (the date is still
debated) had reached the southern tip of
South America, thus populating all the con¬
tinents except Antarctica.
Genetic change did not stop as humans
wandered across the globe. A very large
majority of the genes that define humanity
are shared by all humans, but there are
significant genetic differences between pop¬
ulations. Swedes, Tanzanians, Chinese, and
the Quechua Amerindians who live in the
Andes differ in their appearance - within the
range of human differences - and they differ
in other ways as well. These include sensi¬
tivity to sunlight, the ability to metabolize
milk and milk products, and susceptibility
to a number of diseases. So it is reasonable to
believe that there may be genetic differences
in intelligence as well, and in Chapter 11 we
will review the evidence that there are.
But, and it is a very big but, one of the rea¬
sons that humans were able to settle all over
the globe, in very different ecological sys¬
tems, is that their big brains and enhanced
capabilities for learning allowed them to
203
204
HUMAN INTELLIGENCE
develop different cultures, to adapt to differ¬
ent environments. Then these cultures took
on a life of their own, as they were socially
transmitted from generation to generation.
So it is reasonable to believe that there may
be sociocultural differences in intelligence.
And there are.
Theodosious Dobzhansky, one of the
great geneticists of the twentieth century,
said
Nothing in biology makes sense except in
the light of evolution.
Theodosius Dobzhansky
(The American Biology Teacher,
March 1973, p. 129)
He was right, but when it comes to
human intelligence we have to consider both
biological and social evolution. Is intelli¬
gence something you inherit, genetically,
or acquire through experience? The short
answer is “both.” The long answer is a lot
more complicated.
The bad news is clear. There are genet¬
ically determined conditions that virtually
guarantee mental disability. Mental disabil¬
ity can also be produced by environmental
conditions, notably any event that results in
brain damage. In other cases genetic config¬
urations raise the risk of mental disability,
but the degree of disability depends upon
environmental conditions.
The good news is not so clear. Lots of
things influence intelligence in the normal
range, roughly from IQ 70 upward. Test
scores are not as accurate in predicting cog¬
nitive competence, intelligence in the con¬
ceptual sense, in the upper range as they
are in predicting mental deficiencies. Many
mental disabilities are caused by one or a few
genetic anomalies, or by catastrophic envi¬
ronmental events that produce brain dam¬
age. Variations in intelligence within the
normal range are seldom caused by single
events. They result from the cumulative
effect of many genetic and environmental
factors, with each factor having only a small
effect. Therefore, the effects may be evident
only on a population basis, and can be doc¬
umented only by statistical analyses.
This chapter focuses on the genetic basis
of intelligence. The next discusses environ¬
mental factors. No smoking guns will be
revealed, for there are none. On a popula¬
tion basis, the effects of both genetic inher¬
itance and environmental circumstances are
measurable. Except in the case of men¬
tal disability, assigning a particular person's
intelligence to genetics or environment is
virtually impossible.
The chapter is divided into four sec¬
tions. The first is a quick overview of basic
genetic theory. The second discusses quan¬
titative behavior genetics, a discipline that
is concerned with the extent to which var¬
ious human traits are inherited. We then
look at molecular genetics, which deals with
the mechanisms of genetic inheritance. The
chapter closes with a summarization, and a
discussion of some of the controversies that
have surrounded studies of intelligence and
genetics.
8.2. A Quick Introduction to Genetics
This section is a brief introduction to basic
concepts in genetics. Hopefully it will pro¬
vide an adequate introduction for readers
who have not studied genetics, and a useful
refresher for those who had a course some
time ago.
The genetic model of inheritance was
established in the mid nineteenth century
by the Austrian monk Gregor Mendel (1822-
1884). Today it seems that a monk would
be an unlikely person to make a scien¬
tific contribution, but in Mendel's time a
monk with a scientific career was not that
unusual. Mendel spent a considerable por¬
tion of his professional life at the University
of Vienna, and lectured at other institutions
as well.
Mendel studied the inheritance of eas¬
ily recognizable traits in plants, such as the
color of pea plants. He discovered that a
plant had a genetic potential, derived from
its parent plants, for producing a trait, such
as leaf color or form. Mendel established
that
THE GENETIC BASIS OF INTELLIGENCE
205
(a) The genetic potential of both parents is
passed on to the offspring.
(b) Traits fall into two types. A dominant
trait will be expressed in an offspring
if it has inherited the potential for the
trait from either parent. A recessive trait
is expressed only if it is inherited from
both parents.
(c) Therefore, we have to distinguish
between the observable characteristics
of an individual, the phenotype, and the
genetic configuration, the genotype.
In modern terms, Mendel had inferred
the existence of a gene. He also concluded
that a gene could have two or more different
forms, now called alleles, that might cause
a characteristic to be expressed in different
ways. For instance, a pea’s pod cover can be
either rough or smooth.
Imagine a plant that produces either a
white or a red flower, and that the white
color is dominant. Write C for the allele
that produces white, and c for the allele
that produces red. Furthermore, let the
first term written be the gene inherited
from the “alpha” parent (which in plant
crossings is arbitrary) and the second the
gene inherited from the “beta” parent. We
have the following possibilities, using the
terms homozygous to refer to an individual
whose genetic makeup consists of two
genes with the same allele, and heterozygous
to refer to an individual whose genotype
contains two different alleles.
Genotype
Phenotype
Term
cc
White
Homozygous
Cc
White
Heterozygous
cC
White
Heterozygous
cc
Red
Homozygous
Ignoring whether an allele came from the
alpha or beta parent, there are three possi¬
ble genotypes, CC, Cc, and cc, but only two
possible phenotypes, white and red. The
genotype implies the phenotype; if we know
the genotype of a plant, we know the phe¬
notype. The converse is not always true, for
the phenotype “white flower” could be pro¬
duced by either a CC or a cC genotype.
In simple cases genotypes can be inferred
by combining the information from pheno¬
types of two or more generations of related
individuals. This is called a pedigree study. In
the example, a plant breeder could conduct
a pedigree study by reasoning as follows:
'7/ 1 cross red with red, I will always get
red. "
“ Suppose I cross red with white and,
unknown to me, the white flower has geno¬
type CC. All offspring will have genotype
Cc, and thus I will get only white in the
offspring generation. Let's call this batch of
flowers 2A."
This situation is shown in Figure 8.1(a).
“ But suppose that the white flower has
genotype Cc. Then all the next genera¬
tion will inherit a c from one parent and,
equally likely, either a C or a c from the
other parent. Call this batch of flowers zB.
In the zB population all offspring will have
either genotype Cc or cc, and I expect to
have an equal number of them. "
The resulting distribution of genotypes
and phenotypes will be
Genotype
Probability
Phenotype
Cc
1
2
White
cc
1
2
Red.
The situation
is shown in
Figure 8.1(b)
The frequencies are statistical expecta¬
tions, rather than determined values. There¬
fore, the breeder cannot expect batch 2B
to be made up of an exactly equal num¬
ber of red and white flowers. However, if
just one red is seen, that is evidence that the
white flower in the first generation could not
have been CC. Therefore, the breeder can
be “statistically certain” that the first genera¬
tion white was CC only if the second gener¬
ation is large enough that the probability of
not generating a red from a first generation
Cc white is very low.
2o6
HUMAN INTELLIGENCE
(a)
(b)
Figure 8.1. A simple illustration of Mendelian genetics. The
situations shown are the “pedigrees” that might be derived from
hybridization between two plants bearing a gene that has either a
"white” allele (C), which results in white flowers, or a "red” allele
(c), which results in red flowers. The upper figure, (a), shows the
results that would be expected if a white plant with dominant
genes only (phenotype = white, genotype = CC) is crossed with a
plant carrying recessive genes only (phenotype = red, genotype =
cc). The lower figure shows a similar pedigree assuming that the
white “parent” had genotype Cc (heterozygous).
Pedigree studies provide a way of infer¬
ring genotypes from phenotypes, based on
the pattern of inheritance. The idea works
very well in plants, in situations where the
expression of a phenotype is completely dic¬
tated by the genotype. However, it may
take several generations before the pattern
becomes apparent. This is no problem if we
are dealing with plants that have discrete
phenotypes, such as color or roughness, and
where we know the ancestory of each indi¬
vidual. When it comes to people and intelli¬
gence, neither condition holds. We are deal¬
ing with a continuous trait, so the notion
of discrete phenotypes is blurred. We are
dealing with a relatively slow-breeding pop¬
ulation; a human generation is a little over
thirty years long. In addition, as will be illus¬
trated, some traits may not reveal them¬
selves until adulthood or even old age. We
certainly cannot control mating, so we have
no way of creating individuals with informa¬
tive pedigrees, as a plant or animal breeder
can. Therefore, while pedigree studies are
sometimes useful in human genetics, it is
more often the case that we have to look at
statistics across populations. That turns out
to be an advantage, rather than a defect, for
such studies link genetics to evolution.
8.2.1. Mendel's Model Applied
to Populations
In the 1930s Theodosius Dobzhansky (1900-
1975), a Russian-born biologist who immi¬
grated to the United States in 1927, devel¬
oped techniques for studying genetics at
the population level, without regard for
THE GENETIC BASIS OF INTELLIGENCE
207
individual pedigrees.1 In addition to this
insight, Dobzhansky' s work provided the
essential link between genetics and evolu¬
tion. This section will consider only simple
cases. Further complications are discussed in
subsequent sections.
Dobzhansky worked with the common
fruit fly, Drosophila melanogaster. From a
genetic view, Drosophilia has three attrac¬
tive characteristics: it is cheap; it breeds
rapidly; and it has a number of traits that
are controlled by single genes with only two
alleles, as in the examples mentioned ear¬
lier. Dobzhansky used Mendel’s principles
to develop mathematical models of the dis¬
tribution of traits in successive generations
of a population of fruit flies.
To illustrate, let W and w be the alleles of
the gene controlling the fly's body color. The
normal (“wild type”) body color in fruit flies
is grayish, but there are shiny black (ebony)
colored fruit flies. Imagine that a laboratory
has available two large populations, one of
wild type flies and another of ebony flies.
At this point we do not know the genetic
makeup of either population, because we
do not know if a heterozygous (Ww) fly is
grayish or ebony.
We then conduct an experiment in which
we create a mixed population, containing
70% wild type flies and 30% ebony flies. In
flies it is reasonable to assume random mat¬
ing. The following combinations of parent¬
ages can be expected
Wild with wild: .70 x .70 — .49, just under
half of the flies in the next generation.
Wild male with ebony female: .70 x .30 =
.21, just over one-fifth of the flies.
Wild female with ebony male: .30 x .70 =
.21, just over one-fifth of the flies.
Ebony male with ebony female: .30 x .30 =
.09, just under one-tenth of the flies.
Now suppose that, unknown to us, all
wild flies have genotype WW, all ebony flies
have genotype ww, and W is dominant. If
1 Dobzhansky’s major work, Evolution and the Origin
of the Species, was first published in 1937. Two later
revisions were published, the last in 1951.
these assumptions are correct, 91% (49% +
21% -F 21% = 91%) of the flies in the offspring
generation should have the gray, wild type
phenotype, and 9% should have the ebony
phenotype. The mathematics is a bit more
complicated for the second generation - the
offspring of the offspring - but the same
principle applies. If we start with a mix of
WW and ww populations, then the propor¬
tion of different genotypes, and by implica¬
tion the proportion of phenotypes, is deter¬
mined for all future generations. Statistical
techniques (trust me!) can be used to deter¬
mine the most likely starting configuration,
given the distribution of phenotypes in later
generations.
Similar models can be applied for genes
with more than two alleles and for situa¬
tions in which heterozygous genotypes (Cc
or Ww in the examples) display a phenotype
that is a mix between the two homozygous
forms. Although the models are more com¬
plicated than the example, the principles are
the same.
In order to link genetics to evolution we
need the concept of selective pressure. This
can be illustrated by continuing the exam¬
ple, including the assumption about the par¬
ent population containing 70% WW and 30%
ww genotypes. The distribution of fertilized
fly eggs, not mature flies, will have the fol¬
lowing distribution of genotypes:
Phenotype
Genotype
Fraction
Wild
WW
•49
Wild
Ww
.42
Ebony
ww
.09
It is not immediately obvious, but can
be proven, that if there is random mating
within a population, the relative frequen¬
cies of the genotypes will be maintained.
This is called Hardy -Weinberg equilibrium.
In our example, if all three phenotypes
were equally likely to survive to reproduc¬
tive maturity, the fractions would be main¬
tained in all subsequent generations. If we
were to look ten generations ahead, the frac¬
tions would still be (.49, .42, .09).
2o8
HUMAN INTELLIGENCE
This is where evolution comes in. Sup¬
pose that not all phenotypes are likely to
survive at all, or to produce an equal number
of offspring. In the example, it might be that
ebony-type flies are easier to see, and thus
more subject to predation, than the wild-
type flies. The mechanism of selection does
not matter; the numbers do.
Suppose that the gray-bodied pheno¬
types, all flies with the wild appearance,
have a 90% chance of reaching sexual matu¬
rity (and hence reproducing themselves),
while animals with an ebony-body pheno¬
type have only an 80% chance. This is called
a selection pressure; some genotypes are more
likely to reproduce successfully than are
other genotypes. Selective pressure will dis¬
tort the population frequencies in the breed¬
ing population of the next generation. In this
example, the distribution of genotypes in
the breeding population of the first, second,
and third generations would be:
Generation
WW
Ww
ww
1
•495
.424
.081
2
•5°5
.418
■°77
3
•5*5
•4J3
.073
10
•574
•372
.054.
There is a drift toward domination by
more viable phenotypes (genotypes WW
and Ww), which comprise just under .92
of the genotypes in generation 1, rising to
.946 after ten generations. After 100 genera¬
tions the frequencies would be WW = .851,
Ww = .143, ww = .005, with a .001 rounding
error. The evolutionary pressure favoring a
particular phenotype exerts a pressure that
changes the genotypical distribution in the
population. However, the unfavorable ww
geneotype never quite dies out. Why not?
The w alleles hide out, in the Ww geno¬
type, which has a nonpathological pheno¬
type. The ww form can reoccur when two
Ww genotypes mate. Such matings account
for 99% of the ww (pathological) genotypes
at fifty generations.
Let’s tie this back to human intelligence.
The .9 figure for producing an offspring who
also lives to reproductive age is far more
characteristic of human populations than
of fly populations.2 As the example shows,
selective pressures operate to reduce the fre¬
quency of genes associated with phenotypes
that have low reproductive rates. However,
it may take a long time for the less favorable
genotypes to disappear from the population.
This is illustrated in Figure 8.2, in which
the example has been changed so that the
two phenotypes have reproductive rates of
.9 and .5. The figure shows the resulting fre¬
quencies of genotypes WW, Ww, and ww,
extended over fifty generations. We could
conduct the experiment in flies in about
three months. An analogous experiment in
humans would take over 1,500 years.
While one should always treat hypothet¬
ical examples with reservation, it is interest¬
ing to consider what this implies in social
terms. Typically people know their pheno¬
type, but not their genotype. Imagine a man
and a woman, both of whom are of genotype
Ww. As population figures approach stabil¬
ity, six out of every thousand couples would
fall into this category. Neither would have
any reason to suspect that they were carri¬
ers of the w allele, yet there would be one
chance out of four that their child would
be a ww, and hence exhibit the pathological
phenotype.
Nevertheless, evolution has made things
better. At the start of our hypothetical evo¬
lutionary cycle approximately one in six
couples would have been in this situation.
That is a lot worse than six out of a thou¬
sand in the evolved population.
Could situations like this actually occur
in modern human populations? They can,
and they do. Panel 8.1 presents an example,
Huntington's Disease, and describes a famous
case.
2 The replacement rate is the average number of off¬
spring per woman required to maintain a stable
population. In developed countries this is approxi¬
mately 2.2, which implies that the probability that
an individual offspring will live to reproduce is .909.
THE GENETIC BASIS OF INTELLIGENCE
209
Figure 8.2. The effect of evolutionary selection on the distribution
of genotypes. The relative frequency of three genotypes (ordinate)
is plotted as a function of the number of generations (abscissa) . The
probability that a WW or Ww genotype will live to reproduce is
assumed to be .9. The probability that a ww genotype (leading to a
pathological phenotype) will live to reproduce has been set at .5.
Initially the ww genotype had a frequency of about .05 (one in
twenty cases in the population). By ten generations it had dropped
to .01 (1 in 100). The change from thirty generations (.002) to fifty
generations (.001, 1 in 1,000) was very small.
8.2.2. On beyond Mendel: Complications
when Dealing with Continuous Traits
(like Intelligence)
The single-gene model does a good job of
explaining the inheritance of traits like the
colors and surfaces of pea pods, eye and body
color in fruit flies, and a surprising number
of severe pathologies of human intelligence,
such as Huntington’s disease. Things rapidly
become more complicated for continuous
traits, and for traits that are influenced by
more than one gene. Mendel thought of
some of these complications; others he had
no way of anticipating.
Multiple alleles. A gene can have more
than two alleles. There is an important
case, Alzheimer's dementia, in which this is a
factor.
Additive genetic potentiab for continuous
traits. Alleles do not always fall neatly into
“dominant” and “recessive” categories. The
heterozygous form may have a phenotype
that is intermediate between the phenotypes
of the two homozygous forms. As a hypo¬
thetical example, suppose that the height
of a plant is determined by a single gene,
with two alleles. Represent these by A and
a. If the heterozygous (Aa) genotype has a
potential that is an additive mixture of the
two phenotypes, then we might have the
following situation:
Genotype
Phenotype
Expected Height
AA
HIGH
20 cm
Aa
MEDIUM
15 cm
Aa
LOW
10 cm
Environmental factors complicate the obser¬
vation of genetic and phenotypic variation. A
genotype is associated with the potential for
expression of a trait. Environmental factors
may cause the observed value to vary around
the expected value. In the height example,
suppose that the frequencies of the A and a
forms of the gene were both . 5 and that envi¬
ronmental factors produce deviations from
the genetic expectation that are normally
210
HUMAN INTELLIGENCE
Panel 8.1. Huntington's Disease:
Woody Guthrie and His Family
Huntington’s disease: Huntingon's dis¬
ease is a severe form of mental and
physical disability with an incidence of
approximately 1 case in 30,000 people.
Symptoms usually appear somewhere
between thirty-five and fifty years of
age, although symptoms in adolescence
are not unknown. The initial symptoms
are loss of short-term memory, then
tremors and progressive failure of cogni¬
tion. Abnormal irritation and aggressive
behaviors have also been reported. Loss
of muscular control follows, and then
death .
The disease is caused by a dominant
allele, so affected individuals need have
only one copy of the gene. If a child of
a Huntington's disease patient carries the
allele, he or she is at risk of developing
symptoms earlier than the parent did.
The singer Woodrow Wilson (Woody)
Guthrie (1912-1965) was a prototypical
example of the inheritance of Hunting¬
ton’s disease.
Guthrie was born to poor parents in
Oklahoma shortly before World War I.
His mother, who died in an Oklahoma
mental institution, is now believed to
have suffered from Huntington’s disease.
Guthrie's early adult life was marked by
spectacular success as a folk and social
protest singer in the 1930-40 depression
era. His “This Land is Your Land” is con¬
sidered a classic of the genre. In mid¬
dle age Guthrie developed severe mem¬
ory loss, progressive dementia, and other
signs of deterioration typical of the syn¬
drome. He died of the disease at age fifty-
five.
Guthrie fathered eight children. Three
of them died (two in auto accidents)
at age twenty-one or younger, earlier
than one would expect the disease to
express itself. Therefore, we do not know
if they were carriers of the gene for
Huntington’s disease. Two of the surviv¬
ing children developed Huntington’s dis¬
ease. Both died at forty-one, younger than
their father at the time of his death. The
remaining three, including the contem¬
porary folk singer Arlo Guthrie (1947-),
are apparently not carriers.*
* Information from a website supporting a Public
Broadcasting System tribute to Woody Guthrie,
Wikipedia, and several biographical articles.
distributed with a mean of zero and a stan¬
dard deviation of 2 cm. Figure 8.3(a) shows
the distribution of heights that would result
for each genotype. Figure 8.3(b) shows the
distribution of phenotypes, that is, the dis¬
tribution of observed height. This distribu¬
tion was obtained by summing the distri¬
butions for each genotype, weighted by the
probability of the genotype’s occurrence.
Selection pressure complicates the situa¬
tion. In the example of Figure 8.3, I assumed
that there was no selection pressure. As the
argument that led to Figure 8.2 showed,
differences in probabilities of reproduction
due to selection pressure will affect both
genotype and phenotype frequencies over
time. In the simplest case, which was illus¬
trated in Figure 8.2, there is a “favorable” and
an “unfavorable” allele, and the heterozy¬
gous genotype produces the phenotype, and
therefore the reproductive success, associ¬
ated with the dominant allele. There are
cases in which the phenotype associated
with the heterozygous genotype may have
a reproductive success rate above, interme¬
diate to, or below the success rates of the
phenotypes associated with either homozy¬
gous genotype. Panel 8.2 describes one such
case.
Dominance effects. Dominance refers to
the extent to which one of the alleles in
a heterozygotic pair dominates the other
allele. Introductory examples of Mendel’s
laws usually present examples of complete
dominance, in which the phenotype for the
heterozygotic (Aa) form is the phenotype
THE GENETIC BASIS OF INTELLIGENCE
211
(a)
— aa
- Aa
- AA
Frequency
(b)
Figure 8.3. The distribution of phenotypes with independent
environmental effects. This example shows the effects expected
with three different genotypes, aa; Aa, and AA, with uniform
environmental effects that are independent of the genotypical
potential. See text for details. Figure [a] shows the conditional
distribution of phenotypical values, given the genotype. Figure (b)
shows the overall distribution of the phenotypical values, given
equal frequency of the two alleles.
of the dominant [AA] allele, thus mak¬
ing a sharp distinction between dominant
and recessive alleles. When the phenotypical
trait is continuous, as in the case of intelli¬
gence, the heterozygotic phenotype can be a
compromise between the phenotypes of the
two homozygotic genotypes. That was the
case for the hypothetical plant height illus¬
tration given earlier. In that illustration the
heterozygotic phenotypical height was mid¬
way between the height associated with the
two dominant genotypes. This is not nec¬
essarily the case. For instance, in the plant
height example the Aa genotype might have
been associated with a genotypical height
of 18 cm, closer to the 20 cm phenotypical
height of the AA form.
In statistical analyses dominance effects
appear as interactions, for the phenotypic
effects of the two forms of the allele are not
additive.
Probabilistic display of a phenotypical char¬
acteristic. In some cases genetic makeup
influences the probability of displaying a
trait, rather than the form that the trait has.
This is particularly true of behavioral traits
that depend upon interactions between
the individual and the environment.
212
HUMAN INTELLIGENCE
Panel 8.2. Sickle-Cell Anemia,
Malaria, and the HBB Gene
The HBB gene influences the develop¬
ment ol hemoglobin in the blood. Let H
be the dominant form and h the recessive
form. People who have the hh genotype
suffer from sickle-cell anemia . Untreated,
people with sickle-cell anemia usually die
in their twenties. Treated patients now
live into their thirties and forties.
People who are heterozygous (Hh) for
the HBB gene have sickle-cell trait. The
symptoms of sickle-cell anemia are gen¬
erally not present. However, 25% of the
children of parents with sickle-cell trait
will be homozygous (hh), and therefore
will have sickle-cell anemia. Since there
is a good chance that a child with sickle¬
cell anemia will not live to reproductive
age, especially if not treated, the repro¬
ductive success of the heterozygous par¬
ents is reduced.
To make things more complicated,
the h allele confers some protection
against malaria, a mosquito-borne dis¬
ease. Worldwide, malaria is a major dis¬
ease, especially in developing countries.
There are 355 million to 500 million cases
and about one million deaths each year,
mostly when young children contract the
disease. The disease has been virtually
eradicated in developed countries, largely
due to aggressive measures to control the
mosquito population.
If malaria is prevalent, the order of
reproductive success for the three geno¬
types is Hh, HH, hh. If malaria is not
present, however, the order of reproduc¬
tive success is HH, Hh, hh.*
* Information on sickle-cell anemia retrieved from
www . n 1 m . ni h . gov/ m edl in epl us/ency/article/
0001527.htm. Information on malaria retrieved
fromwww.nlm.nih.gov/medlineplus/ency/artide/
000621.htm, August z8, 2008.
Alcoholism, smoking, and other addictions
illustrate such behaviors. People whose rel¬
atives are addicts are not necessarily slightly
addicted, but the probability that they will
become addicts is increased.
Genes are packaged on chromosomes.
Genes are inherited in packages, rather than
being inherited individually. Every cell in
the body of every living creature contains
structures called chromosomes, which bear
the genes for that individual. The chromo¬
somes come in two types, the autosomes and
the sex chromosomes. In mammals the auto¬
somes are paired, so that each member of a
pair contains one of the alleles of a gene pair.
The number of chromosome pairs varies
with the species. Humans have twenty-two
autosomes, referred to somewhat unimag¬
inatively as chromosomes 1-22. An individ¬
ual inherits one chromosome in each pair
from the mother and the other from the
father.
There are two types of sex chromo¬
somes, X and Y. All normal human females
have two X chromosomes, one inherited
from each parent. Genetic transmission for
females is thus the same on all twenty-
three chromosome pairs. All normal males
have an X chromosome inherited from the
mother and a Y chromosome inherited from
the father. The two chromosomes contain
different genes. Probabilities of inheritance
have to be modified accordingly. This is
shown in the case described in panel 8.3.
We now turn to the case of inheritance
controlled by several genes, the polygenetic
situation.
8.2.3. The Effect of Multiple Genes
(the Polygenic Model)
Suppose that there are two independently
distributed genes, A and B, each with two
equally likely alleles (A, a and B,b), and let
the favorable alleles (A or B) each add five
IQ points above the mean of 100 and the
unfavorable alleles (a or b) each subtract
five IQ points from the mean. There are
THE GENETIC BASIS OF INTELLIGENCE
213
Panel 8.3. Male Pattern Baldness
Male pattern baldness (loss of hair
spreading from just above the temples)
is a sex-linked genetic trait. Male pattern
baldness is not itself a cognitive trait, nor
has it been linked to any variation in intel¬
ligence. The reason for discussing it here
is that it is largely under the control of
genes on the X chromosome. This also
appears to be the case for many cognitive
deficiencies. The ancestral relationships
in male pattern baldness provide an easily
observed analog to ancestral relationships
that may be involved in the inheritance
of some aspects of intelligence. Imag¬
ine a young man concerned over future
hair loss. His father is not bald, but his
two grandfathers both display male pat¬
tern baldness. Should the young man be
worried?
Yes. The risk for male pattern bald¬
ness depends on an allele of the AR gene,
located on the X chromosome.* Because
the young man inherited a Y chromo¬
some, but not an X chromosome, from
his father, the father’s hair pattern is irrel¬
evant. By the same token, the father’s
father (the young man’s paternal grand¬
father) is also irrelevant.
The maternal grandfather is another
matter. The young man inherited his X
chromosome from his mother. Mother
had two X chromosomes, one from the
man’s maternal grandfather and one from
the maternal grandmother. Because the
maternal grandfather has male pattern
baldness, the maternal grandfather prob¬
ably carries the AR allele on his X chro¬
mosome. (There are other causes of bald¬
ness, but we disregard them for sim¬
plicity.) There is a 50% chance that
the maternal grandfather’s gene was car¬
ried forward to the young man, via his
mother. Thus the probability is at least .5
that the young man carries the AR allele,
putting him at risk for baldness.
Why do I say “at least"? There is also
a chance that the AR allele was carried
on the X chromosome that his mother
inherited from her mother (the maternal
grandmother). More generally, if there is
a man with male pattern baldness some¬
where in the maternal line beyond the
grandparents, there is a chance that the
baldness genotype will be carried for¬
ward, unexpressed, through the females
of that line, until it reaches a male. The
chances of this happening grow smaller
and smaller with each generation, but
the chance of inherited baldness never
vanishes.
* Information on male pattern baldness retrieved
from the National Institute of Health website
ghr.nlm.nih.gov, March 29, 2008.
now nine possible genotypes with associated
probabilities and phenotypical intelligence,
which for simplicity I will denote IQ. The
values are shown in Table 8.1.
Several of the genotypes in Table 8.1 pro¬
duce the same phenotype, the same poten¬
tial for an IQ score. Figure 8.4(a) rearranges
the data to show the probability of dif¬
ferent IQ potentials (phenotypes) regard¬
less of the genotype. Figure 8.4(b) does the
same thing for a more complicated case in
which, instead of dealing with two genes,
each of which contributes zb ten IQ points,
we consider a situation in which there are
ten genes, each with two alleles, and each
allele either adds or subtracts .3 points of
potential from the mean of 100. As the
number of genes increases, the distribu¬
tion of IQ potentials approaches the normal
distribution.
8.2.4. A Simple Polygenic Model
of Intelligence
Suppose that genetic and environmen¬
tal effects are additive and independent.
The phenotype, the observable score on a
test, should be predictable by combining
214
HUMAN INTELLIGENCE
Table 8.1. Genotypes and associated IQ potentials for the (artificially simple) case of two
genes affecting intelligence. The assumptions are that each gene has two equally probable
alleles, that a favorable allele contributes 5 points above a mean potential of 100, and that
an unfavorable allele contributes 5 points below the mean potential. The genes are
assumed to be independently distributed.
Genotype aa bb aa Bb Aa bb aa BB Aa Bb AA bb Aa BB AA Bb AA BB
Probability 0.0625 0.1250 0.1250 0.0625 0.2500 0.0625 0.1250 0.1250 0.0625
Expected IQ 80 90 90 100 100 100 110 110 120
genetic and environmental influences, in the
equation
X=G+E+e, (8.1)
where X is the score, G and E are the
genetic and environmental contributions to
the score, and e is a residual term due to test
unreliability.
Equation 8.1 refers to statistical expecta¬
tions and predictions. No one ever achieved
an IQ score because they had such-and-
such genetic inheritance and environmen¬
tal experience. They achieved the score
because of how they behaved on the test,
0.2
a Probability
IQ Potential
(b)
Figure 8.4. Probabilities for IQ potential (ordinate) as a function of the
value of the genotypic potential. Calculations are for a two-gene model
(a) and a ten-gene model (b). See text for further details.
THE GENETIC BASIS OF INTELLIGENCE
215
which was determined by the state of their
brain (and hence mind, for the brain has a
mind of its own) at the moment of testing.
The brain came to be in the state it
was because of the combined influences of
genetic heritage and environmental experi¬
ence. These are distal causes of an IQ score.
The behavior was the proximal cause. When
we start to consider the relative influence
of genetic and environmental influence on
intelligence, the distal-proximal distinction
becomes quite important.
Equation 8.1 cannot be used to break
down the score of any individual into
genetic and environmental components, for
we would be explaining one observable
in terms of two unknowns. The equation
does imply the distribution of scores (and
hence correlations) across individuals of
different degrees of genetic relatedness.
Therefore, we can untangle genetic and
environmental influences by examining vari¬
ations in IQ scores (or other variables) across
individuals.
The variation in a set of scores is mea¬
sured by the standard deviation; the larger
the standard deviation, the more variable
the trait underlying the score. The vari¬
ance is the square of the standard devia¬
tion. The covariance measures the extent to
which two scores vary together. Equation 8.1
implies that
Var(X) = Var(G ) + Var(E)
+ 2 Cov(G, E), (8.2)
where Var indicates variance and Gov indi¬
cates covariance. The term on the left,
[VarX), is the variation of the observed
test scores. The first term on the right,
( VarG ), measures the variation in the
genetic makeup of the population; ( VarE ) is
an index of the variation in the environment;
and the third term, iCov[G, E), reflects the
extent to which genetic and environmental
contributions are correlated.
The extent to which a trait is heritable
is defined as the proportion of variance in
the trait that is due to variance in genetic
contribution. This is referred to as broad
sense heritability,
h2 =
h2 =
Var{G)
1 /ar(X)
(8-3)
Var(G)
Var(G) + Var(E) + 2Cov(Gf E)
This coefficient is the “statistic of choice”
in many discussions of genetic influences on
intelligence.
Heritability is the proportion of phe¬
notypic variance associated with genotypic
variance. It is important to remember (a)
that variance is a measure defined on popu¬
lations, but not on individuals in that pop¬
ulation and (b) that variance is independent
of the value of the expected (mean) value.
These two conclusions also apply to h2. It is
a useful statistic, if we keep certain restric¬
tions in mind.
(a) Unless h 2 is zero or one we do not know
the extent to which any individual's
intelligence, or any other phenotypic
trait, has been determined by heredity
or environment. For instance, modern
studies of hr in European and North
American samples usually result in an
estimated hr of from .5 to .8, depend¬
ing on certain characteristics of the sam¬
ple, such as age. It does not follow that
there is any person in the sample whose
personal intelligence is 50-80% due to
genetic inheritance.
(b) Because the variance is independent of
the mean, an environmental variable
can change the average level of a trait
in that population without influencing
the heritability. A good example is the
height of Japanese- Americans. Height
is a highly heritable trait. In addition,
adult height is strongly dependent on
the quality of infant nutrition. At the
time of the major Japanese immigration
to the United States, in the late nine¬
teenth century, infant nutrition in Japan
was poor by modern standards. A cen¬
tury later Japanese-Americans who are
2l6
HUMAN INTELLIGENCE
entirely descendants of the early immi¬
grants (i.e., who have no non-Japanese
ancestors] are substantially taller than
their forbearers, even though they come
from the same gene pool.
Could the same thing happen with
intelligence? There are indications that
it has. In the twentieth century there
was a marked rise in the IQ scores of
people in the developed nations. This
change; over only three human genera¬
tions; occurred far too quickly to have
been due to genetic changes.
(c) The heritability coefficient is influenced
by three things: the extent of genetic
variation in the population; the pene¬
trance, which is the extent to which the
value of the phenotypic trait is deter¬
mined by the genome; and the extent
of relevant environmental variation in
the population. If a population con¬
sists of genetically diverse individuals;
the heritability coefficient will go up; to
the extent that the population consists
of genetically homogeneous individuals;
the heritability coefficient will go down.
Conversely; if relevant environmental
variation goes up; the heritability coef¬
ficient goes down; and vice versa.
Point (c) is obvious when you inspect
equation 8.3; because the term for genetic
variance appears in both the numera¬
tor and the denominator; while the term
for environmental variance appears only
in the denominator. However; people do
not always think mathematically; especially
when the discussion of the genetics of intel¬
ligence becomes heated, so let us examine a
few cases.
In the extreme; an army of clones (as in
the science fiction Star Wars films) would
have no genetic variability; and hence the
heritability coefficient would be zero. This
extreme is never reached in human popu¬
lations; but different populations do vary a
good deal in the extent of their genetic vari¬
ability. Sub-Saharan African populations
display considerably greater genetic varia¬
tion than other populations. The genetic
variability in sub-Saharan Africa should
drive the heritability coefficient upward
compared to genetic variability in other large
populations.
Small; isolated populations generally
show low genetic variability. The inhabi¬
tants of Norfolk Island in the Pacific Ocean
provide a historically interesting case. Vir¬
tually all the modern inhabitants of Nor¬
folk Island are descended from nine British
sailors and twelve Polynesian women who
fled to nearby Pitcairn Island after the
famous Bounty mutiny in 1789. (Their
descendents moved to richer; and then unin¬
habited; Norfolk Island in 1856.) We would
expect the heritability coefficient to be low
in the present-day Norfolk Islanders; due to
low genetic variability.
Heritability also depends upon the extent
to which there is environmental variation
relevant to intelligence. Increases in relevant
environmental variation decrease heritabil¬
ity; decreases in relevant environmental vari¬
ation increase heritability. To illustrate the
point I offer a thought experiment.
I conjecture that h2 in the population
of England has gone up since the nine¬
teenth century, Gabon's time. Why? For
simplicity; let us disregard immigration;
and assume that all present English citi¬
zens are descended from the English pop¬
ulation of the nineteenth century. As a
first approximation; assume that the vari¬
ance in the genetic contribution to intel¬
ligence; Var(G); is the same today as it
was in the nineteenth century. As Dickens’s
novels so graphically portray; in the nine¬
teenth century the economic; educational;
and health differences between the rich and
poor were stark. In the England of today rea¬
sonable health care; education; and nutrition
are available to nearly everyone. In terms
of equation 8.3; the variance in the envi¬
ronmental influence on intelligence; in the
denominator of the equation; has decreased.
Therefore; if the genetic variation is con¬
stant the heritability measure; h 2 , will have
gone up.
My idea about heritability in nineteenth-
century England was labeled a conjec¬
ture because I do not have measures of
either genetic or environmental variation
THE GENETIC BASIS OF INTELLIGENCE
217
in nineteenth-century England. However, a
modern study, that had half the data, sup¬
ports the conjecture.
In modern European-North American
industrial and post-industrial populations,
estimates of h2 fall in the range .40 < h2 <
.60. This value may not hold in subpopu¬
lations. Studies in the United States have
found h2 values in the .5-6 range for school-
age children whose parents are of middle
to high socioeconomic status (SES), but
values of less than .40 for the children of
low SES parents.3 The investigators sug¬
gested that this is because variations in envi¬
ronment above a threshold quality, which
most middle and high SES families exceed,
have relatively little effect on intelligence.
However, variations in environmental qual¬
ity below this threshold, which may occur
in many low SES families, have a substan¬
tial effect on intelligence. This is a reason¬
able explanation of the findings, and there is
some evidence, based upon direct observa¬
tion of the differences in children’s environ¬
ments in high and low SES homes, that the
hypothesized environmental differences do
exist.4
Gene x environment correlations are
expressed by the Cov(G, E] term in
equation 8.2. It is possible that there are
genetic and environmental influences that,
causally, have separate influences on a phe¬
notypic trait, including intelligence, and
whose distributions are correlated in the
population. Because the Cov(G, E] term is
in the denominator in equation 8.3, if the
correlation between genetic and environ¬
mental effects goes up, hr will come down.
That is simple algebra. The causal relations
are more interesting, for at times discussions
of the inheritance of intelligence have been
faulted by a failure to consider gene x envi¬
ronment correlations. The issue has been
particularly important when conclusions are
based upon an observation of a correlation
between a person's intelligence and par¬
ental SES.
3 Harden, Turkheimer, & Loehlin, 2007; Turkheimer
et al., 2003.
4 See Nisbett, 2009, for a review of this evidence.
Gabon’s Hereditary Genius, one of the
seminal nineteenth-century books on intel¬
ligence, is almost a prototypical example of a
failure to consider gene-environment corre¬
lations. Galton examined historical records
of “eminent men,” ranging from members
of Parliament to war heroes, and found that
they came from a relatively few families. He
concluded:
The direct result of this inquiry is to make
manifest the great and measurable differ¬
ences between the mental and bodily facil¬
ities of individuals, and to prove that the
laws of heredity are as applicable to the
former as the latter. Its indirect result is to
show that a vast but unused power is vested
in each generation over the very nature of
their successors-that is, over their inborn
facilities and dispositions.
Francis Galton, Hereditary
Genius, Introduction
to the 1892 reprinting
Galton ignored the possibility that great
families have social and economic resources
that further their children’s careers in ways
that have nothing to do with biology. Could
this happen today? What do you think? To
assist you, panel 8.4 provides an American
example that may be food for thought.
Galton leaped to a genetic conclusion,
ignoring the effects of gene x environment
correlations. Let us look at a contemporary
error, but in the opposite direction.
One of the charges made against the use
of the SAT as a college entrance require¬
ment is that because examinees from high
SES families obtain high test scores, the
SAT perpetuates the “unfair” advantage that
these examinees have because of their social
background.5 This argument turns Gabon’s
conclusion on its head; the observation that
there is a correlation between parental SES
and SAT scores is taken as proof that chil¬
dren from well-to-do families have an unfair
advantage because family wealth has been
used to construct good environments for
learning, including coaching on how to take
the test. Therefore, the critics aver, the SAT
should not be used to screen applicants to
5 Lemann, 1999.
2l8
HUMAN INTELLIGENCE
Panel 8.4. Has Anyone Ever
Inherited the American Presidency?
As of 2010 the United States had had
forty-five presidents. Four American fam¬
ilies contributed two presidents apiece:
two father-son combinations (John and
John Quincy Adams, George H. W.
and George W. Bush), one grandfather-
grandson combination (William and Ben¬
jamin Harrison), and two distant cousins
(Theodore and Franklin Roosevelt). The
probability of this occurring by chance
is miniscule; even in the Adams’ time
the population of the United States num¬
bered in the millions. Does this mean that
the United States never completely shook
off the genetic determinism exemplified
by the British royal family?
We should not arbitrarily rule out
the possibility that genetics contributed
to these men’s ability to behave in a
way that would ultimately bring them to
the presidency. But let us consider some
gene by environment correlations. It is
well documented that the elder Adams
and Bush assisted their sons in their
early careers. (An unfriendly newspaper
columnist described George W. Bush as
having been born with a silver shoe in
his mouth.) Theodore and Franklin Roo¬
sevelt were born to separate branches
of the very rich Roosevelt family. They
both suffered severe illnesses (Theodore
in childhood, Franklin as an adult) that,
in less affluent individuals, would have
severely restricted their career prospects.
Each Roosevelt received the best treat¬
ment that the medical science of the
day could offer, and price was not an
issue.
So was ascension to the presidency
dominated by genetics or environment?
It is impossible to say.
college, because doing so will solidify class
lines.6 This conclusion could be right, but it
could be that relatively wealthy, highly edu¬
cated parents pass on a genetic advantage
to their children. If this is the reason that
children from higher SES families get better
SAT scores, then a different argument has
to be made about the use or nonuse of the
test.
In fairness to the many investigators who
have conducted research in this area, I has¬
ten to add that it is not easy to measure
gene x environment correlations. The con¬
ceptual issue is easy to grasp. In practice,
though, we have a model for calculating
genetic resemblances, but we do not have
any such model for calculating environmen¬
tal resemblances. Here is the problem.
Mendelian genetics provides a model for
the genotypic resemblance between two
individuals with specified family relation¬
ships. You obtain half your genetic material
from your mother and half from your father;
6 Atkinson, 2005.
going back a bit, you obtain a quarter of your
genetic material from each grandparent. We
do not have any such model to specify envi¬
ronmental relationships. The difference in
genetic distance between grandparents, par¬
ents, and self is clear. The difference in envi¬
ronmental distance between generations is
not clear at all.
As a result, investigators either treat envi¬
ronmental effects as an unmeasured resid¬
ual - that is, a “left over” effect - or they try
to measure environmental resemblances by
determining the correlations between cog¬
nitive measures and proxy measures of the
environment, such as parental education,
income, or, in one case I have seen, gov¬
ernment ratings of a neighborhood as being
“poor,” “middle-class,” “upper middle-class,”
or “rich.” In many studies only one such vari¬
able (e.g., parental education) is measured.
There is a good chance that some crucial part
of the environment will be overlooked. The
size of the effect associated with the missing
part can be measured, but the nature of the
effect remains unknown.
THE GENETIC BASIS OF INTELLIGENCE
219
8.2.5. Looking inside Genetic Effects
Equations 8. 1-8. 3 treated genetic effects as
a single package, G. Conceptually, G is the
sum of four components;
G=A+D+I + {G:E) (8.4)
where A refers to additive genetic effects,
D refers to dominance effects, I refers to
a phenomenon called epistasis, and (G : E)
refers to gene-environment interactions. Let
us take each of these in turn.
A is the sum of the independent effects
of the genes. The D term adjusts for domi¬
nance, the fact that the influence of two alle¬
les of the same gene may not be strictly addi¬
tive. In a gene with two alleles, dominance
occurs if the trait expressed by the heterozy-
gotic case does not represent an average
of the expressions of the two homozygotic
cases.
Epistasis, the I term, refers to interactive
effects between different genes.7 Epistasis
occurs if the expression of one gene depends
upon alleles present in another gene. To take
an oversimplified example from evolution,
the presence of feathers does not guarantee
flight, nor does the presence of light bone
structure in the forelimbs. If you combine
them, and make a few other adjustments to
the limb structure, none of which alone per¬
mits flight, you have a bird!
As this example illustrates, on the evo¬
lutionary scale epistatic effects have been
spectacular. Within the field of human intel¬
ligence, though, epistatic effects have vir¬
tually been ignored. This may be a mis¬
take, for as we gain a better idea of how
the genes work we are seeing more and
more cases in which one part of the genome
regulates the expression of another part.
At present, though, we are only begin¬
ning to identify the genes involved in intel¬
ligence. When these are known, epistatic
effects may come to play a greater part in
our thinking about intelligence than they
do now.
7 Plomin et alv 2008, p. 377.
The (G : E)8 term refers to gene-environ¬
ment interactions. A gene-environment
interaction occurs when an environmental
variable alters the expression of a genetic
potential, or when the genetic character¬
istic changes the influence of an envi¬
ronmental variable. Alcoholism is a good
example. Protracted alcoholism can have
disastrous effects on cognition. Susceptibil¬
ity to alcoholism has a substantial genetic
component.9 In societies in which alcohol is
widely used as a recreational drug, a person
who has an inherited susceptibility to alco¬
hol is at risk for cognitive deficit. There is
no genetic risk in a society that does not use
alcohol.
Gene-environment interactions are not
the same as gene-environment correlations.
A gene-environment correlation occurs
when genetic and environmental influences
make contributions to a trait that are inde¬
pendent within an individual, but have cor¬
related distributions across the population.
For instance, the argument against Galton’s
conclusion about genius, and the extent to
which parental SES contributes to a per¬
son’s SAT score, are examples of gene-
environment correlations.
With a few exceptions (notably in
the work of Urie Bronfrenbrenner and
Stephen Ceci of Cornell University),10 gene-
environment interactions relevant to intelli¬
gence have not received very much atten¬
tion. Part of the problem is a technical one.
It is difficult to separate gene-environment
interactions and gene-environment correla¬
tions by statistical analysis. Hopefully this
will change as better models are developed.11
There is also the continuing problem of
deciding what environmental variables to
measure. In some cases, though, the prob¬
lem is conceptual. The alcoholism example
was clear-cut. But what happens when the
8 This term is often written G x E. I have chosen a
different notation to avoid confusion between the
genetic concept of interaction and the statistical
concept, which depends upon the multiplication
operation.
9 Plomin et al., 2008, pp. 270-277.
10 Bronfrenbrenner & Ceci, 1994; Ceci, 1990.
11 Johnson (2007) provides a good candidate for such
a model.
220
HUMAN INTELLIGENCE
interaction is social? Consider the case of
children’s intelligence.
Children's IQ scores are correlated with
parental SES, with values in the .^-.4. range.
Why? One possibility is that the effect is
entirely environmental. Higher SES parents
spend more time interacting with their chil¬
dren in ways that develop cognition than do
lower SES parents.12 For instance, the high
SES parent is more likely to encourage chil¬
dren to solve problems on their own than is
the low SES parent. The physical environ¬
ment is also better in the high SES home;
the high SES home provides more books
and creative toys. It could be argued that
this is an example of a gene-environment
correlation; the lower SES children would
have higher IQs if they had the home advan¬
tages of the high SES children. It may be,
though, that there are genetic influences.
High SES parents may be genetically pre¬
disposed to interact with their children in
particular manners, and the high SES child
may have inherited a genetic potential to be
receptive to these interactions.
This example shows how difficult it is
to untangle interaction effects when we are
comparing a distal and a proximal influence.
The genotype does not influence IQ scores
in the same sense that it influences eye color.
The IQ score is determined by the exami¬
nee's behavior in a testing situation, which
is in turn determined by the examinee’s
experiences and ability to learn from them.
However, people influence the way oth¬
ers react to them. Therefore, the genotype
exerts a distal influence on IQ scores, while
experiences exert proximal influences. Add
to this the fact that there are correlations
between environmental opportunity and
genetic constitution, and it is easy to see why
sophisticated statistical models are needed
to separate main effects, interactions, and
correlations.
8.2.6. Calculating Genetic Variance
Quantitative behavior genetics (QBG)
attempts to identify the extent to which
12 Nisbett (2009) presents this argument in detail.
variations in human traits are associated
with genetic or environmental influences.
In this section I will try to explain how this
is done. I will start with a simplified model
that illustrates the basic principles, and then
explain the more realistic ACE model. First
a few words are in order about the QBG
approach itself.
QBG deals with statistical associations
between variation in genetic and environ¬
mental makeup and variation in displays
of intelligence, usually test scores. This
is necessary because it is impossible to
test causal hypotheses by controlling either
genetic makeup or environmental condi¬
tions in humans, in the way that might
be done in studies of plants or nonhuman
animals.
Probably the easiest examples of QBG
to understand are analyses of similarities
and differences between monozygotic (MZ,
“identical”) and dizygotic (DZ, “fraternal”)
twins. Monozygotic (MZ) twins result from
the splitting of a single fertilized ovum into
two genetically identical ova, while dizy¬
gotic (DZ) twins result from the fertiliza¬
tion of two ova during a single act of sexual
intercourse. This provides a “natural exper¬
iment,” in which the MZ twins are genet¬
ically identical, while the DZ twins, like
siblings (SS), share half of the permissible
human variation in genetic material. 13 The
difference is important.
The correlation between scholastic
achievement test scores of ten-year-old
identical (monozygotic, MZ) twins is about
.78, varying a little with the academic topic.
The correlation for fraternal (dizygotic,
DZ) twins is about .50.14 Intuitively the
higher correlation in MZ twins suggests that
genetic similarity is associated with similar
academic accomplishments.
13 Statements that two individuals share a certain per¬
centage of their genes are a bit misleading. Most of
the genes in the human genome are shared with all
animals, and many are shared with plants. Other
genes are invariant over all people. These are the
genes that define the species. The statement that
two individuals share x% of their genes refers to the
extent to which the genotypes of two individuals
covary over those genes that can vary in humans.
14 Plomin et al.; 2008, Table 9.5.
THE GENETIC BASIS OF INTELLIGENCE
221
Figure 8.5. The relationship between genetic
and environmental contributions in a single
individual.
To go further, we need a more precise
model of how the observed correlations are
produced. Modem QBG relies heavily on
a statistical technique called path analysis
or, somewhat more frequently, structural
equation modeling. The factor analytic tech¬
niques used in psychometrics (Chapter 4)
are special cases of structural equation mod¬
eling. Historically, though, the factor ana¬
lytic techniques were developed first.
Here is an illustrative simple model. Sup¬
pose that the genetic and environmental
components are independent of each other.
The phenotype for person i can be expres¬
sed as
X2 = hxGi +exEit (8.5)
where X,- is person is score on test x (or
any other record of a trait), Gz is Vs genetic
potential, and E: refers to environmental
effects that are statistically independent of,
and hence unpredictable by, genetic poten¬
tial. This model is shown in Figure 8.5.
The goal of the analysis is to determine
the values of h and e, the coefficients repre¬
senting the strength of the genetic and envi¬
ronmental influences. These are called the
path coefficients. We cannot measure any¬
thing but X directly, so we cannot compute
a simple regression of X on G and E. What
we can observe, however, are the correla¬
tions (covariances) between the X values for
people of known relationship. The variance
in a trait can be broken down into its com¬
ponents;
Var(X) = h2Var[G) + e2Var[E)
+ iheCov[G, E ) (8.6)
where Var indicates a variance and Cov
a covariance. In words, equation 8.6 says
that the variance in test scores is equal to
the variance in genetic potential, weighted
by the square of the genetic path coeffi¬
cient, plus the variance in the environmen¬
tal potential, weighted by the square of the
environmental path coefficient, plus a term
that depends upon the two path coefficients
and the covariance between the genetic and
environmental potentials.
Since the scales of all variables are arbi¬
trary, we can deal with standardized vari¬
ables, all of which have a mean of zero
and a standard deviation of 1. Equation 8.6
becomes
1 ~ h2 + e2 + 2 her[G, £), (8.7)
where r(G, E) refers to the correlation
between genetic and environmental poten¬
tials.
If we assume that there are no gene-
environment correlations, r(G, E] = o, equa¬
tion 8.7 becomes
1- h2+e2. (8.8)
The parameter h2 can be interpreted as
the percent of variance in the phenotypi¬
cal trait (X) that is associated with additive
genetic effects. In general, a path coefficient
is associated with every arc in the graphic
version of the structural equation model. If
there is no arc between two entities, as is the
case for the G and E entities in Figure 8.5, a
path coefficient of o is implied.
The next step is to make use of the fact
that we know the degree of genetic rela¬
tionship between any pair of relatives - for
example, between siblings or between chil¬
dren and grandchildren. This lets us expand
the basic model, as shown graphically in
Figure 8.6. Instead of considering a single
person, i, we consider two people, i and
i', who stand in some known relation to
each other - for example, MZ or DZ twins.
Obviously there are equivalent equations
and diagrams for each person individually.
Figure 8.6 puts the two figures together,
as indicated by the double-headed arrows,
222
HUMAN INTELLIGENCE
Figure 8.6. A model of the relation between genetic and
environmental effects in two individuals, i and i'. Known path
parameters are shown in boldface. If the kinship relation between
the individuals is known (sibs, first cousins, etc.), the genetic
correlation, r(Gj, Gj ), is known from genetic theory. The
correlation between scores on trait X, r(Xj, Xj), can be observed.
Other parameters must be estimated. If one can assume that the
environments are independent, then the environmental correlation
r(Ei, Ej ) can be set to zero.
which denote correlations between the two
people's environments, r[Eu £/), genetic
constitutions, r(Gt; G/], and scores on the
trait in question, e.g. IQ scores, r(XI; X/).
The path coefficients and correlations,
collectively, constitute the observables and
the parameters of the model. The observ¬
able, shown in boldface in Figure 8.6, is the
correlation between test scores. The param¬
eters are either fixed or free. Fixed param¬
eters (also shown in boldface in the fig¬
ure) are parameters that are not directly
observable, but are fixed by theoretical
considerations. This is the point where
QBG uses genetic theory. The correlation
between genetic potentials is a fixed param¬
eter, because genetic theory specifies the
fraction of genetic variation shared by two
related people. This can vary from i, for
monozygotic (MZ) twins, to o for genet¬
ically unrelated individuals. The h and e
coefficients, which represent the effects of
heredity and environment, are free parame¬
ters. Path analysis is used to find values for
the free parameters that, when combined
with the fixed parameters, reconstruct the
correlations between observables as closely
as possible.
It can be proven, although I will not do
so here/5 that the correlation between two
observable variables will be equal to the
sum of the products of the coefficients along
each of the paths linking them in the dia¬
gram. Therefore, in this model the correla¬
tion between the X and Xf scores is calcu¬
lated across pairs of people of a given degree
of relationship, such as all MZ or all DZ twin
pairs.
r[X, X') = hxr(G, G)hx + exr(E, E')ex
r{X,X’) = hlr[G,G) + elr(E,E')
(8.9)
Equation 8.9 expresses one observable,
r(X, X) in terms of four values on the
15 See Bollen, 1989, or Loehlin, 2004, for discussions.
THE GENETIC BASIS OF INTELLIGENCE
223
right-hand side. Such equations are mathe¬
matically indeterminate, because more than
one value for the right-hand side variables
could be chosen to satisfy the equation.
(This is generally the case whenever there
are more values to be calculated than observ¬
ables.) The solution to the problem is to con¬
sider different degrees of genetic and envi¬
ronmental relationships simultaneously.
As an example, suppose that we study
two samples - one of same-sex siblings raised
in their parents’ home, and one of same-sex
siblings in which each sibling is adopted into
a different home. We make the following
assumptions:
1. The adoptions occur very early in
the children’s life, ideally shortly after
birth.
2. The adoption agency does not prac¬
tice selective placement. There was no
attempt to match the characteristics of
the adoptive and biological parents.
Under these assumptions it is reasonable
to assume that the environmental corre¬
lation between the adopted siblings, r(£,
E ), is zero.16 The equation for the cor¬
relation between two individuals adopted
into different homes, equation (8.9), simpli¬
fies to
rapan{X, X) = /i2r(G, G). (8.10)
At this point QBG introduces genetic
theory. Genetic theory tells us that for DZ
twins and same-sex siblings, r(G, G) = .5.
The next step is to introduce observ¬
ables. As quite a few studies of siblings
raised together and apart have been con¬
ducted, we have a good idea of the value
of the correlation between test scores,
r(X, X), for siblings raised together or
apart - raised together .47, raised apart .24. 17
16 This assumption is widely made. However, it can be
challenged. Nisbett (2009) has speculated that the
very fact that two families are willing to adopt may
make home environments similar in ways that are
important to cognition.
17 Plomin et al., 2008, Figure 8.7.
Therefore, for the siblings raised apart
r apart [X, X') = fcj(.5) (8.1l)
h2 = rgpUK X')
•5
K = -48-
For siblings raised together there will be
some unknown but possibly nonzero envi¬
ronmental correlation, r[E , E ), reflecting the
fact that two children raised in the same
family are likely to have similar environ¬
ments:
rtogetheriX, X') = ^(.5) + eV(E, £').
(8.12)
Making appropriate substitutions for
observed and theoretically established val¬
ues,
h2x (.5) = .24 from equation 8.11
e2 = 1 — h2 from equation 8.8 (8.13)
h2x (,5) + ezr[E, E ') from equation 8.12.
This is a system of three linear equations
in three variables, and hence has a unique
solution. Rearranging the first line estab¬
lishes that h 2 = .48. Substituting this value
into the second line, e2 = .52. Using these
values to solve for the remaining equation,
we have r[E, Er) = .44.
Conceptually, the model tells us that
if our assumptions are correct the variance
in the test scores is approximately equally
associated with genetic variance and vari¬
ance in the environment that is statistically
independent of genetic variance, and that
siblings raised together experience statisti¬
cally related, but not identical, environmen¬
tal influences.
But how do we tell if our assumptions are
correct?
8.2.7. Fitting Models to Data
The example just presented illustrates a
saturated model. In saturated models only
224
HUMAN INTELLIGENCE
one set of parameters will satisfy the data,
and these parameters can be used to recon¬
struct the observed matrix of correlations or
covariances perfectly. While studying satu¬
rated models can be informative, the fact
that the parameters inevitably reconstruct
the observations means that there has been
no reduction in the complexity of our expla¬
nation. To illustrate, in the example just
given only two parameters, Fandr(£, F),
were estimated from the data, because the
value of e 2 was dictated once h2 was deter¬
mined. Therefore, we explained two obser¬
vations, r foggier [X, X') and rapan{X,X), in
terms of two estimated parameters. What
we would like to do is to provide an expla¬
nation that is less complex than the orig¬
inal observations, by deriving K parame¬
ters to explain M observations, where K is
(substantially) less than M. In order to do
this we have to relax our requirement that
the parameters reconstruct the observations
exactly. Instead, we try to find parameters
that can be used to derive expected observa¬
tions that can be compared to the observed
values.
This is done using complicated statisti¬
cal estimation methods that are far beyond
the scope of this book. However, a simple
illustration can be used to show the spirit of
model fitting, without going into computa¬
tional details.
Suppose that we add to the previ¬
ous example the case of dizygotic (DZ)
twins raised together and apart. In the¬
ory, this should not make any difference,
because DZ twins share the same amount
of genetic material as do biological same-
sex siblings (SS). In fact, though, the
correlations are raised; r DZ{together)[X, X') =
.55 compared to rSs(togetker) (V X') = .47,
and rDx(apart](X, X') = .35 compared to
^SS(apart) ( X, X*} = .24*
If we repeat the analysis of equations 8-13,
using DZ correlations instead of SS correla¬
tions, our estimates become h 2 = .70, e 2 =
.30, andr(£, F) = .67 (actually 2/3). These
numbers reconstruct the DZ twins data,
but do not reconstruct the data for siblings.
What we want are parameter estimates that
can be used to approximate both the SS and
the DZ data. We could average the values
of the estimates obtained from SS and DZ
cases, so that
h‘ _ 7° + ^ ^
rtE.E-) = feL±!=.555.
The observed values of the correlations
between test scores now differ from their
predicted values:
Correlation
Predicted Value
Observed Value
^SSapart
■295
.24
^SStogether
.522
•47
r DZapart
.295
■35
rDZtogether
.522
■55
We have obtained greater parsimony, by
reconstructing four observations from two
estimated parameters, at the expense of
accuracy.
At this point a certain amount of judg¬
ment is used. One could argue that genetics
and environment must work the same way in
both DZ and SS pairs, but because DZ pairs
are the same age, their environments may
be more similar than the environments of
SS pairs. This is not an entirely ad hoc argu¬
ment, for SS pairs are, by definition, made
up of an older and a younger sibling, and
just being the elder or the younger may have
different influences on development. If you
accept this argument, you should require
that h 2 have the same value for DZ and SS
pairs, but you allow for separate values for
rss(£, F) and roz[E, F). This changes the
predicted-observed comparison to
Correlation Predicted Value Observed Value
^SSapart
.295
.24
^SStogether
.476
■47
^DZapart
.295
•35
f"DZtogether
.568
•55
The predicted and observed values are
now closer to each other than before, but
some parsimony has been lost, because four
THE GENETIC BASIS OF INTELLIGENCE
225
observations are being derived from three
estimates.
This example mirrors the use of structural
equation modeling in modern QBG and, for
that matter, in other fields in the behav¬
ioral and social sciences. The only major
difference between what has been shown
and what is done in practice is in the way
the parameters are estimated. The averag¬
ing method was used solely because it is
easy to understand. In practice scientists use
computer- intensive techniques of parame¬
ter estimation, which we need not discuss.
The important thing is to understand the
principles behind model fitting, rather than
the computational details.
8.2.8. A Rough-and-Ready Method
for Estimating h2
Prior to the development of structural equa¬
tion modeling QBG relied heavily on “rough
and ready” methods for calculating h2. This
often forced researchers to rely on question¬
able assumptions. Here are three commonly
offered comparisons
One estimation method is based on com¬
parisons of the correlations between scores
of MZ and DZ twins. If the twins are raised
in the same family,
tmz = h2 -\-e2r{E, E ')
rDZ = ±/P + e2r(E,F) (8.14)
h1 = 2 (rMz - tdz )■
This contrast is valid if the correlation
between environmental influences is the
same for pairs of MZ and (same-sex) DZ
twins. If MZ twins live in more similar envi¬
ronments than DZ twins, the approximation
in equation 8.14 will overestimate the heri-
tability coefficient.
A similar logic can be used to contrast
the MZ twin correlations to (same-sex) sib¬
ling correlations, or to parent-child corre¬
lations. In both cases the assumption that
the same environmental correlation applies
across types of relationship is suspect.
Another contrast is based on the test
scores of MZ and DZ twin pairs when both
twin pairs have been raised apart. Such sit¬
uations are unusual, but they do occur. If
it can be assumed that the adoption pro¬
cess itself does not introduce a correlation
between environments, the environmental
correlation, r[E, F), will be zero for both
MZ and DZ twins. Therefore,
rMz(X,X')= h 2
(8.15)
rDZ(X,X) = i/P.
8.2.9. The ACE Model
We now examine a realistic, widely used
model in QBG research on intelligence,
the ACE model. This model assumes three
sources of variation in test scores or similar
indices:
A: Additive genetic effects. The sum of the
effects of different genes, without consider¬
ing dominance and epistatic effects.
C: Shared environmental effects. “ Shared "
here refers to shared variation. Recall that
QBG models operate on pairs of individu¬
als, such as siblings or twins, who stand in
some relation to each other. Shared envi¬
ronmental effects refer to variations in the
environment that vary across pairs, but
are shared within a pair. For siblings, the
clearest example ofnonshared environment
effects is a difference in the family environ¬
ment. Even if both members of a sibling
pair grow up in the same family, one will
be older than the other, so the older and
younger sib have different family environ¬
ments.
E: Nonshared environmental effects. These
are effects of the environment that differ
both across pairs and within a pair. Con¬
tinuing with the sibling example, each sib¬
ling is part of the other sibling’s entnron-
ment. An environmental effect associated
with being an older or younger sibling -
and there are such effects, as we shall see
in Chapter 9 - would be a nonshared envi¬
ronmental effect.
The causal equation for the ACE model
is
X- A+C+E,
(8.16)
226
HUMAN INTELLIGENCE
Figure 8.7. The ACE model. The path coefficient r(A, A ) is
determined by genetic theory. The path coefficient r{C, C) is set to
1, by the definition of shared environmental effects that act on each
member of a pair. By a similar definition, r{E, El ) must be zero, and
therefore is not contained in the model. The correlation between
observable traits, r(X, X'), is determined by the data. Genetic and
environmental coefficients a , c, e are estimated by fitting the model
to the data for three or more types of pairs (e.g., MZ and DZ twins
and siblings}.
where X is a phenotypical trait score, A
refers to additive genetic effects, C refers to
shared environmental effects, and E refers to
nonshared environmental effects. The path
diagram for the ACE model is shown in
Figure 8.7. As in Figure 8.6, observable corre¬
lations and theoretically dictated parameters
are shown in boldface, and free parameters
are shown in open type. The model per¬
mits estimation of heritability coefficients
(h2) and separate environmental coefficients
for shared (c2) and nonshared (V) environ¬
mental influences.
The ACE model is the model of choice
for many of today's QBG studies of cog¬
nitive and personality traits. It has been
criticized for failing to allow for gene-by¬
environment interactions and correlations.
If these are present they will be erroneously
assigned either to environmental or genetic
main effects. Alternative models that do
permit measurement of interactions have
recently been developed/8 but they require
more observed correlations than the ACE
model requires.
18 Johnson, 2007.
8.3. Observed Estimates of Heritability
Table 8.2 summarizes the results of over
200 studies of the heritability of intelli¬
gence, conducted up to 2000. I know of
no studies since then that would substan¬
tially alter these results. Most of these stud¬
ies involved European or North American,
usually White, participants, with sampling
somewhat but not markedly biased against
the inclusion of low SES participants. In
theory, studies in other populations could
produce different results. However, stud¬
ies done in both urban and rural India
and in Japan have produced comparable
correlations.19
19 These studies all rely on a comparison of the corre¬
lations between monozygotic and dyzygotic twins
reared together. The Indian estimates, based on
two small studies, report MZ correlations in the .9
range, which is exceptionally high for this sort of
study. The DZ correlations were slightly less than
.50. The resulting heritability estimate was in the
.80 range (Nathawat & Puri, 1995; Pal, Shyam, &
Singh, 1997). The Japanese study (Lynn & Hattori,
1990) produced a heritability estimate of approxi¬
mately .60. Indian results are close to the highest
values reported for studies using a contrast between
MZ and DZ twins, but they are not entirely out of
THE GENETIC BASIS OF INTELLIGENCE
227
Table 8.2. Correlations and heritability estimates for pairs of individuals of various
degrees of relationship. The right-hand column shows the heritability coefficients
calculated using the simple model of Figure 8.5, which assumes that environmental
correlations are zero for individuals not raised in the same home.
Relationship
Raised
in Same
Home ?
Degree
of Genetic
Relationship
Correlation
of Test
Scores
h 2 Estimate
Comparison
Used to Make
Estimate
MZ twins
Yes
1
.86
*
MZ twins
No
1
.76
.76
MZ apart
DZ twins
Yes
-5
•55
.60
2(MZ
together-DZ
together)
DZ twins
No
•5
•35
(0, 82,
0)-7°
(1) 2 (MZ
apart-DZ apart)
(2) 2 (DZ apart)
Siblings (SS) raised
together
Yes
•5
•47
.78
2(MZ
together-sibling
together)
Siblings (SS) raised
apart
No
•5
.24
.48
2 (SS apart)
Unrelated individuals
of different ages, raised
in same homes;
Intelligence measured
in childhood
Yes
0
■25
0
*
“Virtual twins”
measured in childhood
Yes
0
.26
*
*
Unrelated individuals
of different age, raised
in same homes;
Intelligence measured
in adulthood
Yes
0
.02
*
*
Parent-child
Yes
•5
•42
*
*
Parent-child
No
•5
•H
.48
2 (Parent-child
apart]
Note: The data is based on summaries by Scarr, McGue et al., and by Plomin et al. and virtual twin
studies by N. Segal and her colleagues (McGue et al., 1993; Plomin et al., 1997, p. 140; Scarr, 1997,
p. 28]. Unless specified otherwise the studies summarized were based on scores obtained when the
offspring were in childhood and adolescence.
The right-hand column of Table 8.2 is the
estimate of heritability obtained using the
simple model presented in Figure 8.5, which
line. The Japanese study is in agreement with esti¬
mates of heritability obtained in Western industri¬
alized nations.
assumes independent and additive environ¬
mental and genetic effects. Granted that
the model is simplified, let us consider the
important trends.
Heritability estimates lie in the .5-. 8
range.
228
HUMAN INTELLIGENCE
There are environmental effects. For all
four nonzero biological relationships the
correlation between pairs is from .10 to .20
points higher when the pairs are raised in
the same home. This increase is in the cor¬
relation coefficient, and is not affected by
changes in the level of scores. We will deal
with this in Chapter 9, where adoption
effects are discussed.
The three rows labeled “unrelated indi¬
viduals raised in the same household” and
“virtual twins” are of special interest. “Same
household” correlations refer either to
children and adopted children in the same
family, or to unrelated individuals who
have been adopted into the same family.
“Virtual twins” are a subset of the “same
household” group in which the individuals
are within nine months of age of each other,
and have lived in the same household since
before their first birthday. The correlation
between cognitive test scores of virtual
twins is .26 when taken in childhood. It
would be of considerable interest to know if
this correlation changes over the life span,
but as of this writing the necessary research
has not been done.
While there is considerable variability in
the estimates, the variability is systematic.
Heritability estimates involving twins are
higher than heritability estimates that do not
involve twins. This seems to be a fairly con¬
sistent finding over the studies I have exam¬
ined. The correlation between the scores of
DZ twins raised apart is substantially higher
than the correlation between the scores
of siblings raised apart, even though their
genetic relationships are identical. Why? We
do not know, but can consider some possi¬
bilities.
The method of estimation assumed that
there were no gene-by-environment interac¬
tions. To the extent that a person’s genetic
makeup is important in shaping the envi¬
ronment, omitting this factor would inflate
the correlation between MZ twins. A similar
factor could influence the contrast between
DZ twins and siblings. DZ and SS pairs
are equal in the extent to which they share
genetic makeup. However, DZ twins share
the same prenatal environment, while SS
pairs do not. Once they are born, any time-
locked phenomenon that might influence
cognition, such as exposure to measles or
hepatitis, will influence DZ twins at the
same age (which in some cases is an index
of vulnerability), while SS pairs, being of
different ages, will have different degrees of
vulnerability. In the case of children raised
together, SS pairs exert asymmetric influ¬
ences on each other’s environment, because
one will be older than the other.
The rough and ready estimates of heri¬
tability presented in Table 8.2 cannot be dis¬
regarded. However, the exact values should
not be regarded as set in stone, due to
the failure to consider gene-by-environment
interactions and correlations.
8.3.1. Adoption Studies
Adoption studies provide a way of separat¬
ing environmental and genetic influences on
intelligence. Environmental effects are indi¬
cated by correlations between an adoptee
and members of his or her adoptive family,
genetic influences by correlations between
the intelligence of an adoptee and the bio¬
logical parents. In the ideal adoption study
the children will have been adopted very
early in life, data will be available on the
intelligence of both biological parents, and
the adoptees will be followed until adult¬
hood, to allow for dissipation of environ¬
mental effects and the effect of genetic traits
that become apparent during adolescence,
adulthood, and possibly old age. Such an
ideal study has not been done, but several
have come close. Four adoption studies are
described in panel 8.5.
The results of virtually all adoption stud¬
ies are consistent with the simple environ¬
ment + genetic model described in Fig¬
ures 8.5 and 8.6. As one reviewer put it,
The heritability of cognitive ability is
around .50 when collapsing across all stud¬
ies. However heritability appears to vary
with age, with h2 = .40 in early childhood,
rising to .60 in early adulthood, finally ris¬
ing to h2 = .80 in later life.
Petrill, 2002, p. 284
THE GENETIC BASIS OF INTELLIGENCE
229
Panel 8.5. Adoption Projects
Studies comparing adopted children to
children living with their biological par¬
ents provide an opportunity to compare
the effects of genetic inheritance and
environmental influences on intelligence.
The analysis is not simple.
At first glance, it might seem that
any influence on an adoptee’s intelli¬
gence that was associated with the adop¬
tive family would be independent of the
adoptee’s genotype. This is not correct,
for to a considerable extent people make
their own environments. This can be
apparent at a very young age. Children’s
reactivity has a partially genetic basis, and
reactivity will influence how adults and
other children interact with the adoptee.
Such effects cannot be captured by a
statistical model that assumes the non¬
existence of gene-by-environment inter¬
actions. Unfortunately, while it is
easy to conceptualize a model that
includes gene-by-environment interac¬
tions, it can be very difficult to obtain suf¬
ficient measurements to evaluate such a
model.
Then we come to the practical rea¬
sons. In order to achieve adequate statis¬
tical reliability, and in order to allow for
drop-outs, a good study should involve
at least 200 people, and preferably more.
The studies are of most value if the
children can be followed to maturity.
Because of the difficulty of following peo¬
ple lor that long, it is desirable to take
as many measures as possible. (While we
will concentrate on measures of intelli¬
gence, many adoption studies also involve
detailed studies of personality character¬
istics.] Large-scale studies that take place
over an extended period of time require
substantial amounts of long-term fund¬
ing, which is difficult to obtain.
From the point of view of a study
designer, adoption should occur as soon
after birth as possible, in order to mini¬
mize the exposure that adoptees have to
an environment related to the biological
parents. The study designer wants place¬
ments to be random, so that there is no
correlation between genetic and environ¬
mental potential. In practice, this may
not happen, especially if the adoption
agency attempts to match characteristics
of the biological parents (or, more likely,
the biological mother) with characteris¬
tics of the adopting family. Such a policy
would produce a gene-environment cor¬
relation that would be hard to evaluate.
Generalization is a problem. Children
put up for adoption and the families
who wish to adopt them are not repre¬
sentative of the general population. Par¬
ents who wish to adopt a child tend to
be wealthier and better-educated than
parents who put their children up for
adoption. There is further self-selection
in transracial adoptions and in adoptions
of foreign children. Adoption agencies,
quite reasonably, try to ensure that the
child will be raised in a socially and eco¬
nomically healthy environment. This is a
reasonable policy, but it does restrict the
range of environments to which adopted
children are exposed. The very fact that
families are willing to adopt children indi¬
cates that, as a group, they place consid¬
erable value on raising children. All these
effects act to reduce the variance in the
environment of adopting families, com¬
pared to the range of environmental vari¬
ables in the general population. There¬
fore, adoption studies may underestimate
the extent to which environmental vari¬
ables influence the development of intel¬
ligence in the population at large.*
This panel describes four adoption
studies carried out in the United States.
They are somewhat different, but the
findings are generally compatible. Taken
together, they provide valuable infor¬
mation about the relative influences
of genetic inheritance and environment
upon intelligence.
(continued)
23°
HUMAN INTELLIGENCE
Panel 8.5 ( continued )
The Texas Adoption Project
The Texas Adoption Project' began in the
early 1960s. It focused on 300 adoptees
from a church-affiliated adoption home
for unwed mothers. Both the biological
mothers and the adoptive parents were,
for the most part, white and middle-class.
This probably reflects the social mores
of the day. A woman who was a single
parent faced greater social disapproval in
the 1960s than she would today, and as a
result there was more pressure to give up
a baby for adoption.
The adoptees, their biological moth¬
ers, adopting parents, and the biolog¬
ical children of adopting-parents were
given intelligence tests. Either the WAIS
or a nonverbal test was given, partly
because of the varying ages of the chil¬
dren involved and also, in the case of the
biological mothers, because of adminis¬
trative conditions in the adoption home.
The biological parents had an average IQ
score of 100, but there was less variability
in test scores than would be the case in
the general population. The adoptive par¬
ents’ IQs were slightly higher than those
of the biological parents IQs - 100-101 for
mothers, depending on the test used, and
104 or 105 for fathers. The mean IQ scores
of the natural and adopted children were
in the 104-105 range, again depending on
what test was used. The restriction in
range in all groups would have the effect
of reducing correlations between the test
scores of various classes of individuals,
compared to the equivalent correlations
in a representative population. The study
is also different from several other stud¬
ies in that the biological mothers and
the adoptive parents had roughly equiva¬
lent SES. This contrasts with many other
studies where adoptive parents tend to
be both better-educated and economi¬
cally better-off (and hence better able to
afford to raise a child) than the biological
mothers.
There were two waves of testing, an
initial wave when the adopted children
were eight years old, on the average (with
a considerable range), and a follow-up
of about 50% of the original sample ten
years later. A variety of cognitive, per¬
sonality, and health measures were taken.
Twenty years after the initiation of the
study, a postal questionnaire was sent
to all adoptees and biological children
of participants and to the adoptive par¬
ents. The questionnaire covered various
aspects of life adjustment, such as occu¬
pational and marital status.
Genetic influences were substantially
larger than environmental influences.
The correlations between cognitive test
scores of the birth mother and the
adopted child's scores ranged from .23 to
.36 In the initial study, depending upon
the intelligence test used, and from .26
to .78 (only a small number of cases)
at the ten-year follow-up. The change
over time illustrates Petrill’s comment
that the heritability coefficient increases
with time. The correlations between
scores of biologically unrelated individu¬
als — adoptive mothers and fathers, and
adopted children - ranged from .19 to
.08 (again, depending on the test used
and whether the mother or father’s score
is being considered) at the original test¬
ing, and from .15 to —.02 in the follow¬
up.
The postal survey twenty years later
showed that the biological children of the
adopting parents were doing slightly bet¬
ter than the adopted children, at least
in terms of conventional, self-reported
social adjustment. Adopting-parent char¬
acteristics were better predictors of the
social adjustment of their biological than
of their adopted children.
THE GENETIC BASIS OF INTELLIGENCE
231
The Colorado Adoption Study
The Colorado Adoption Study, begun in
the 1970s, was a twenty-year study of chil¬
dren recruited from two adoption agen¬
cies in Denver, Colorado.* As was the
case for the Texas study, the participants
were largely White and middle-class. In
addition to studying the development
of adopted children, the experiment
included a comparison group of over 200
families consisting of biological parents
and children. Cognitive tests were con¬
ducted when the children were one, two,
three, four, seven, twelve, and sixteen
years of age. This made it possible to
compare stability of intelligence, in the
sense of the relative standing of a person
compared to others of the same age, from
infancy to adolescence.
The correlations between the test
scores of adoptees and their adoptive par¬
ents were near zero at all ages. The cor¬
relations between the biological mothers’
scores and the adoptees’ scores increased
from .18 in early and middle childhood
to .38 at age sixteen. Factor analyses of
the test scores showed that general intel¬
ligence, as indicated by the first factor
on the cognitive tests, g, was responsible
for the consistency in individual perfor¬
mance, and that this factor had a sub¬
stantial genetic load. Instabilities in test
scores over time appeared to be related to
differences in nonshared, within-family,
environmental influences.
The University of Minnesota Studies
The next two studies were both done at
the University of Minnesota/1 The trans-
racial adoption study (TRA) dealt with
African American children adopted by
White families. Unlike the Texas and
Colorado studies, in the TRA there was
a substantial disparity between the socio¬
economic status (and concomitant educa¬
tional levels and intelligence test scores)
of the biological mother and the adop¬
tive parents. This difference was central
to the study, which was motivated by a
desire to see what the effects of improved
environmental conditions would be on
the adopted children. The issues involv¬
ing racial differences in intelligence will
be discussed in Chapter 11. The corre¬
lation between scores for adoptive par¬
ents and adopted children was .29, while
the correlation for biological mothers and
adopted children was .43. These correla¬
tions are somewhat higher but not out of
line with the general trends reported in
Table 8.2.
The adolescent adoption (AA) study
was a study of people ranging in age
from sixteen to twenty- two. The inves¬
tigators tested a group of adoptees who
had spent an average of eighteen years
in their adopted homes. The correlation
between scores of biologically related
siblings was .35, while the correlation
between adoptees and unrelated siblings,
raised in the same home for an average of
eighteen years, was zero!
Virtually all adoption studies in North
America and Europe produce similar
results. The pattern of correlations indi¬
cates that h for general intelligence is
at least .50. Environmental effects due to
differences between families can be sub¬
stantial early in childhood, but decrease
in adolescence and almost vanish in adult¬
hood. This suggests a dissipation in the
effects of the adopting home over a per¬
son’s lifetime, which is certainly reason¬
able. The genetic inheritance is retained
throughout life.
* Nisbett, 2009.
* Loehlin, Horn, & Willerman, 1997; Loehlin,
Horn, & Ernst, 2007.
1 Petrill et al., 2004; Plomin et al., 1997.
$ Scarr & Weinberg, 1983.
HUMAN INTELLIGENCE
232
Pe trill’s summary, and others like it, are
statements about the effect of adoption
upon variation in measures of cognition.
They say nothing about the effect of adop¬
tion upon the mean scores of adopted chil¬
dren, that is, about any benefit or general
cost of adoption. Such effects are found.
There is generally a temporary elevation of
cognitive skills, as indicated by test scores,
through the school years, but this dissipates
in adolescence and adulthood. This does
not mean that adoption does no good, for
an increase in cognitive abilities during the
school years can itself be important to devel¬
opment later in life.
The second qualification has to do with
generalization. Can heritability coefficients
from studies of adopted children be extrap¬
olated to people in general? A case can be
made that they cannot, on the grounds that
adults who adopt children are less variable,
within their own population, than parents
in general. For example, the median age of
women who adopt children is thirty. This
is higher than the median age of birth for
women who raise their own children. Adults
who adopt tend to be of higher SES and
to have more education than typical par¬
ents. These tendencies work to reduce the
environmental variability among adopting
families to a lower value than population
variability. As pointed out earlier, if envi¬
ronmental variance is reduced, heritability
coefficients will go up.20 The extent of the
upward bias in heritability introduced by the
homogeneity of adopting families is hard to
estimate, but the fact that there is some bias
should be kept in mind.
8.3.2. Turin Studies
The study of twins is of great interest in
QBG, because the contrast between MZ and
DZ twins shows the effects associated with
a clearly established degree of genetic dif¬
ference, between people of the same age,
raised in either the same or different envi¬
ronments. Doing such studies is difficult, but
not impossible, due to the relative scarcity
20 Stoolmiller, 1999.
of twins. DZ twins occur about once in 100
births, MZ twins once in 250 births.21
The results from different studies are
consistent.22 The MZ correlations are about
.8, regardless of the twins’ age. The correla¬
tions between DZ twins are about .6 until
adulthood, at which point they drop to .4.
This implies that the relative importance of
genetic influences on intelligence increases
in adulthood. The extent of the increase is
illustrated in Figure 8.8, which shows the
results of a large study in the Netherlands,
which is typical of other findings.23 The heri¬
tability coefficient was modest in early child¬
hood, but increased to a staggering .85 in late
adulthood. The result is similar to results
found in a US study of twins.24
While the findings from these studies are
consistent among themselves, recent stud¬
ies have questioned the generality of their
conclusions. Most of the twin studies rely
on voluntary participation, sometimes over
a considerable period of time. This intro¬
duces a bias toward the participation of
better-educated, higher socioeconomic sta¬
tus children. Two recent studies of very large
samples, taken over a wider range of the
population, illustrate how this bias may
affect estimates of the heritability of cog¬
nitive ability. One of these studies utilized
WISC scores for seven-year-old children,
obtained as part of a national study of chil¬
dren's health. The overall sample of over
300 twin pairs was split into a low SES group
(one-third of the mothers in this group were
on welfare) and a high SES group. The
differences were dramatic. The heritability
coefficient, h2, was .72 in the high SES group
and only .10 in the low SES group. A second
study found a similar pattern in seventeen-
year-olds.25
A similar study from the Netherlands
found little, if any, difference in the
21 The figures are given for natural births for Whites
in the United States. African-Americans give birth
to more twins, and Asian- Americans to fewer.
22 McGue et al., 1993, p. 63. See also Boomsma,
Busjahn, & Peltonen, 2002.
23 Posthuma, De Geus, & Boomsma, 2004.
24 McGue et al., 1993, p. 72.
25 Turkheimer et al., 2003; Harden et al., 2007.
THE GENETIC BASIS OF INTELLIGENCE
233
■ Additive genetic
□ Shared environment
□ Nonshared environment
0 20 40 60 80 100
Percent variance accounted for
Figure 8.8. Percentage of the variance in intelligence test score
account for by additive genetic influence, shared environmental
influence, and nonshared environmental influence. Based on
selected data from the Dutch Twins Study; Posthuma, De Geus,
& Boomsma, 2004, Figure 9.1.
heritability of adult intelligence as a func¬
tion of various socioeconomic indicators.26
The authors of that study suggest that the
expression of intelligence in children may be
more sensitive to variations in the environ¬
ment than is the expression of intelligence
in adults. The authors also point out that
studies of variation, which is typically the
target of a QBG analysis, do not evaluate dif¬
ferences in means across groups. The study
authors found a positive association between
SES and adult intelligence test scores apart
from the influence of heritability.
With a few exceptions, the twins studied
in the research just described were raised
in the same home, by their biological par¬
ents. Conceptually, the almost-perfect study
of the heritability of IQ is the study of the
intelligence of MZ and DZ twins raised in
different households by adoptive parents,
preferably without any interaction between
the two twins. Pragmatically, such a study
is hard to achieve. Twins are rare enough,
and only a small fraction of all twins are
adopted apart. The first study of similarities
in MZ twins raised apart was a case study
of a single pair, in 1922. There were only
three reputable studies, with twelve, nine¬
teen, and thirty-eight pairs of twins, in the
next forty years.27 In spite of the small sam¬
ple sizes these studies produced consistent
results. The correlations between test scores
26 Van der Sluis, Willemsen, et al., 2008.
27 Bouchard, 1997, Table 5.1 and accompanying text.
for MZ twins raised apart ( Tmza ) were, for
each study, .71, .69, and .75.
There was also a larger, highly publicized
study that reported tuza = -77* Unfortu¬
nately, in one of the more embarrassing
moments in the history of Psychology, it
was subsequently found that the data was
not credible. Whether this was due to fraud
or to unacceptably careless record keeping
is still debated. The incident is described in
panel 8.6.
The situation today is much better.
Beginning in the 1970s, Thomas Bouchard,
Jr., and his colleagues at the University of
Minnesota began the Minnesota Study of
Twins Raised Apart (MISTRA). Details
of MISTRA are provided in panel 8.7. When
enrollment ceased, in 2000, the MISTRA
researchers had evaluated 139 twin pairs, in
addition to evaluating the twins’ spouses,
partners, and some adoptive and biological
relatives. The evaluations included person¬
ality testing and collection of biographical
and anthropometric data. In the intelligence
part of the evaluation participants took three
previously developed batteries of cognitive
tests, totaling forty-two tests of cognitive
skills.28
About the same time that MISTRA
began, a group of Swedish researchers ini¬
tiated an extensive study of twins, raised
together and raised apart.29 Sweden, like
28 See Johnson et al., 2007a, or Johnson & Bouchard,
2005, for a detailed listing of the tests.
29 Finkel & Pedersen, 2004.
HUMAN INTELLIGENCE
234
Panel 8.6. The History of Studies
of Twins Adopted Apart: I. Sir
Cyril Burt
This panel describes one of the fascinat¬
ing stories in the history of psychology,
complete with intrigue and, perhaps, a
villain.
Cyril Burt (1883-1971) was an ex¬
tremely influential British psychologist.
He was an early factor analyst, claiming
credit for some of the major advances
in the field. He held important posts
in educational policy and research and
conducted important research on juve¬
nile delinquency. In 1932 Burt succeeded
Spearman as Professor of Psychology at
University College, London. A Professor¬
ship was then equivalent to a lifetime
appointment as a Department Chair.
The appointment carried with it much
more authority and prestige than do
such positions today. He appears to have
discharged his duties in an intellectu¬
ally impressive but socially imperious
manner.
Burt retired in 1950 but continued to
publish until his death in 1971. Among
his other honors Burt was knighted and
received a gold medal for distinguished
lifetime contribution from the American
Psychological Association.
Sometime before World War II, Burt
began collecting data on twins. He pub¬
lished a variety of papers on the topic in
the interval between 1943 and 1966. An
additional paper was published posthu¬
mously in 1972. The most noted of these
papers dealt with MZ twins raised apart.
Some of the correlations reported were
in the high .8’s and even .9’s, indicat¬
ing almost complete heritability of intel¬
ligence. Related papers confirming Burt’s
results were published by a "Miss Con¬
way" and a “Miss Howard.”
In 1974, three years after Burt’s death,
Leon Kamin, a professor at Princeton
and an ardent opponent of the idea
that intelligence is determined geneti¬
cally, observed that some of Burt’s cor¬
relations, which were purportedly on
different data sets, were identical to the
third decimal point. The probability of
this happening by chance is miniscule.
Subsequently, Arthur Jensen, who is an
advocate of genetic theories of intelli¬
gence, made a searching reexamination
of Burt’s data, and found additional sus¬
picious entries.* Jensen pointed out that
the results could either have been fraud¬
ulent or perhaps the result of confusion,
as Burt was elderly at the time the papers
were written. The plot thickened when a
biographer of Burt’s, Leslie Hearnshaw,
found papers that suggested both that
some of Burt's data was fraudulent and
that the papers by Conway and Howard
had in fact been written by Burt himself. ^
There appear to be few, if any, records
that unambiguously document Conway’s
or Howard's career. Since Jensen’s paper,
in 1974, no knowledgeable scientist cites
Burt’s data.
I do not think we will ever know
whether Burt was intentionally fraudu¬
lent or unacceptably careless in his later
years. The cases for and against him rest
largely on circumstantial evidence. But
the issue is not whether Burt was a fraud
or a good scientist gone bad as he aged.
It is whether or not he hindered the
advancement of knowledge.
The correlations Burt reported are
only slightly higher than those reported,
with far better data and no hint of fraud,
by several contemporary studies of twins
raised apart. So in the narrow sense Burt
did not harm science. His claims did not
lead scientists up a garden path, chasing
a false result.
In a broader sense, Burt did profound
harm to science in general, to Psychology,
and in particular to Behavior Genetics.
In the 1960s and 1970s psychologists
and educators emphasized the role of
learning and social opportunity as agents
for the development of cognitive skills.
In the United States these beliefs were
THE GENETIC BASIS OF INTELLIGENCE
235
closely tied to social and moral issues,
for there were great hopes that the end
of segregated schooling would quickly
result in economic and social equality
between African Americans and Whites.
Similar feelings were also strongly held
in Europe. However, progress toward
equality following desegregation proved
to be much slower than had been
anticipated.
In 1969 Jensen published a highly con¬
troversial paper in the Harvard Edu¬
cational Review in which he pointed
out that Black- White differences in aca¬
demic achievement could be due in
part to genetic differences between
the races.* Jensen cited ten of Burt’s
papers in support of his argument.
Jensen s article elicited a furious counter¬
argument. The subsequent revelation
that Burt’s data was, at best, incompetent
and, at worst, fraudulent led not only to
a rejection of Jensen’s argument but also
to a general, albeit unjustified, condem¬
nation of genetic studies of intelligence.
This hurt the field in a variety of ways,
not the least of which was a drying up of
funds for research.
Science ought to inform policy mak¬
ers wrestling with political and social
issues. The only claim that scientists can
make to a privileged status in such a
discussion is their commitment to sup¬
porting their arguments with objectively
collected data and thoughtful analysis,
rather than moral suasion. When a sci¬
entist offers an opinion based on indefen¬
sible data it hurts us all. ^
My comments are based on a book on Burt
edited by Macintosh (1995) that includes papers
by two who had known Burt personally, Arthur
Jensen and Hans Eysenck.
* Jensen, 1974.
Hearnshaw, 1979.
5 Jensen, 1969.
§ See Hunt & Carlson, 2oo7a,br for elaborations
on this point.
many European countries, maintains gov¬
ernment records of the health of its citi¬
zens. In 1978 some researchers noticed that
a substantial number of twins born prior
to 1945 indicated that they had been raised
separately as children.50 The investigators
contacted all twins who were age fifty or
older, and many of them agreed to inter¬
views and fairly extensive psychological test¬
ing, although far short of the testing involved
in the MISTRA project.
The Swedish and MISTRA studies com¬
plement each other. MISTRA dealt with
twins aged eighteen upward into their six¬
ties and seventies, in a single testing, using
a cross-sectional design. The Swedish study
tested participants on four different occa¬
sions, spanning a thirteen-year period. Com¬
paring the studies, the MISTRA study cov-
30 These twins would have been at least thirty-three
years old. They all would have been born either
during the worldwide depression or during World
War II. Sweden was neutral in the war, and received
substantial numbers of refugee children from neigh¬
boring countries.
ered a wider age range, while the Swedish
study had relatively more participants in the
fifty-plus age range. The MISTRA study had
considerably more data on any one individ¬
ual than did the Swedish study. Because of
its longitudinal design, the Swedish study
provided an opportunity to look at changes
in cognition at the time of life when some
people show marked cognitive decline.
The MISTRA data has been ana¬
lyzed using Johnson and Bouchard's g-VPR
model,51 producing heritability estimates for
g and the lower-stratum abilities, includ¬
ing the broad VPR factors. In addition, the
MISTRA investigators were able to deter¬
mine whether or not the different abili¬
ties are influenced by the same genetic fac¬
tors. This is determined by calculating the
genetic correlation, which can be thought of
as an estimate of the correlation between the
genetic influences on each of the two abili¬
ties being compared. The idea is shown, in
diagrammatic form, in Figure 8.9.
31 Johnson et al., 2007.
236
HUMAN INTELLIGENCE
Panel 8.7. Studies of Twins Adopted
Apart: II. The Minnesota Study of
Twins Raised Apart (MISTRA)
Cyril Burt's emphasis on the study of MZ
and DZ twins reared apart (panel 8.6}
was a good idea, whatever the demerits of
his own research. Both before and since
Burt's time several studies of MZ and DZ
twins have been attempted, but only a
few of them have involved a substantial
number of twins reared apart. The rea¬
son is simple: when one combines the
criterion “being an MZ twin” and “being
adopted away from your birth partner”
you do not have very many people
left.
In 1979 a group at the University of
Minnesota, under the general leadership
of Thomas Bouchard, Jr., embarked on
an attempt to find as many twins raised
apart as they possibly could. Recruitment
was by advertising and word-of-mouth.
By the time recruitment was terminated,
in 2000, 139 twin pairs had been enrolled.
In addition to the twins, Bouchard and
his group interviewed and tested as many
related people, including spouses, as they
could find. Eventually over 400 people
participated. The testing took approxi¬
mately fifty hours, over a week-long visit
to Minneapolis, and included personality
tests, physical measurements, and collec¬
tion of biographic data as well as exten¬
sive intelligence testing. The result has
been the compilation of one of the most
important and extensive databases on
twins raised apart that exists.
Bouchard and his colleagues deserve
great credit for their perseverance. Espe¬
cially in the early years, government grant
money for this important research was
hard to come by, in no small part because
of the political reaction against the idea
that there are genetic influences on intel¬
ligence (see panel 8.6}. The situation was
not helped by the questionable status
of Burt’s widely discussed studies, but
I think that the reaction would have
occurred anyway, due to deeply held
political/social beliefs that disparities in
education and socioeconomic status in
the United States are almost entirely due
to environmental causes, often associated
with past prejudice and privilege. Obvi¬
ously I cannot prove this. Bouchard has
expressed similar concerns.*
In addition to the investigators, the
University of Minnesota deserves credit
for its willingness to support Bouchard
and his group in this important, extended
research effort.
* Bouchard, 1997, pp. 126-127.
Figure 8.10 shows heritabilities [h2] cal¬
culated for the latent traits at each
stratum. The figure shows the division
between genetic and environmental associa¬
tions within each of the latent traits at each
level - that is, the g level, the three VPR
dimensions at the third level, and the nar¬
rower cognitive skills at the second level.
With the exception of the perceptual trait,
every third- and fourth-level trait has a her-
itability greater than .40. General intelli¬
gence (g) has a heritability value of more
than .70. The genetic correlations between
traits were substantial, showing that
related genetic influences contribute to all
abilities.
Similar results were found in the Swedish
study, where the investigators used the
scores provided by the WAIS rather than
the g-VPR model. The heritability estimate
for the g factor was .91! This is the highest
heritability estimate for intelligence that I
have seen, in any substantial study.^2
8.3.3. The Unfolding of Cognitive Abilities
Turning back to Figure 8.8, we can see that in
the Dutch study there was a regular increase
in the heritability coefficient from childhood
to age fifty. We see the same trend across
32 Reynolds et al., 2005, Table 5.
THE GENETIC BASIS OF INTELLIGENCE
237
Figure 8.9. A schematic of genetic correlations
involving the P and R factors in the VPR model
of intelligence. The genetic correlation,
r(Gp, Gr), is a measure of the extent to which
the genes responsible for each trait covary with
each other. Genetic correlations can vary
independently of the genetic loadings.
Therefore, two traits might have high
herit ability and a low genetic correlation, or a
low (but nonzero) heritability and a high genetic
correlation.
studies. In the Colorado Adoption Study
the heritability coefficient was .56 at age
sixteen;33 in the Swedish study it was .91 for
people age fifty and older. At first this defies
logic, for, as Bouchard put it to me in pri¬
vate correspondence, one would think that
the slings and arrows of outrageous fortune
would accumulate, producing environmen¬
tal differences over the years. Are there clues
in the data indicating why this result might
occur?
I am struck by two pieces of evidence.
The first is that the heritability estimate is
influenced by the difference between twin
correlations in MZ and DZ twins; the larger
the difference, the larger the heritability
estimate. In a review of the data available up
to 1989, 34 it was clear that the increase in the
heritability estimates was due to a decrease
in the DZ correlation from .6 (childhood) to
.4 (adult) while the MZ correlations stayed
constant at slightly above .8. The second
piece of evidence comes from the Swedish
study of twins over fifty. Recall that in
that study cognitive abilities were measured
several times, making it possible to assess
33 Plomin et al., 1997.
34 McGue et al., 1993, Figure 1.
both the overall cognitive level of partic¬
ipants over the fifty- to- eighty interval and
the change in cognitive abilities over that
interval. As would be expected, there was a
general decrease in cognition as the partici¬
pants moved into their sixties and seventies.
Structural equation modeling showed that
this decrease was largely due to nonshared
environmental differences, the differences
between twins, rather than the environmen¬
tal differences across twin pairs. Note that
“nonshared environmental differences,, has
a rather different interpretation when the
comparison is based on adults living apart,
rather than on children living in the same
household.
I suggest that there are two processes
going on. One is that some genetically influ¬
enced biological processes present them¬
selves only during adulthood and, especially,
old age. Huntington’s disease (panel 8.1)
and incipient signs of Alzheimer’s Demen¬
tia, are examples. Among the healthy
elderly, genetically influenced limitations
on information-processing capacities may
become more constraining as we grow older.
These phenomena may account for a sub¬
stantial part of the increase in heritability.
This sort of genetic limit would not
be surprising from an evolutionary point
of view. Genetic weaknesses that are not
expressed until after the reproductive years
do not affect a person’s reproductive fitness,
so there would be no selective pressure on
the genes involved.
The second process is quite different. It is
suggested to me by the constancy of the MZ
correlations as the DZ correlations fall. To
a substantial degree, people make their own
environments. Genetic influences, acting as
distal causes, influence exposure to environ¬
ments that, acting as proximal causes, dif¬
ferentially accelerate or decelerate cognitive
change. DZ twins, being genetically more
distinct than MZ twins, will select environ¬
ments that differ more than the environ¬
ments selected by MZ twins, and, as a result,
the DZ correlations will go down over time,
due to nonshared (between twins) environ¬
mental differences being greater in the DZ
than in the MZ twins.
238
HUMAN INTELLIGENCE
Figure 8.10. The relative proportions of variance associated with
genetic and environmental variance in the factors defined for the
g-VPR model. Data based on Johnson et al., 2007, Table 5.
This analysis is a conjecture. Proof will
have to wait for further research.
8.3.4. The Heritability of
Information-processing Traits
While there have been studies of the genetic
basis of a large number of information¬
processing traits, our interest centers on
the two information-processing functions
shown to be most closely related to intelli¬
gence (Chapter 6), speed of processing and
working memory. For the reasons given in
Chapter 6, I will be concerned with work¬
ing memory, overall, and will not attempt to
fractionate it into finer components.
Both working memory and processing
speed show substantial genetic heritability.
The genetic correlations indicate that the
genetic basis of the information-processing
components is the same or nearly the same
as the genetic basis of general intelligence.
Most of our knowledge of this comes from
studies of twins. Four such studies will be
described. They have been selected to make
a point about consistency of results.
The first of the studies was of Dutch
twins, born in the early 1990s and evaluated
at five and twelve years of age.55 Siblings
35 Polderman et al., 2006, 2007. See Polderman et al.,
2007, for a discussion of previous work on this topic.
were also studied. Participants took a com¬
prehensive battery of markers for psycho¬
metric g and tasks evaluating the speed and
capacity of working memory. Heritability
estimates for both speed of processing and
the capacity of working memory were in
the .5-. 6 range, increasing slightly from age
five to age twelve. Stability, in the sense
of the extent to which scores at age five
could predict subsequent scores, appeared
to be largely mediated by genetic influences.
The authors point out that this is impres¬
sive, because the brain undergoes substan¬
tial development from five to twelve years
of age.
An Australian study56 evaluated genetic
contributions to information processing and
intelligence test scores in sixteen-year-old
twins and their siblings. Heritability esti¬
mates for choice reaction time tasks var¬
ied from .7 to .5, with heritability increas¬
ing as the number of choices increased. A
delayed reaction time task was used to eval¬
uate working memory. It had a heritability
estimate of .48.
In Japan a study was conducted of young
adult twins (aged fourteen to twenty-nine,
mean age twenty). The twins were given
an intelligence test and both verbal and
spatial-visual working memory tasks. The
36 Luciano, Wright, et al., 2001.
THE GENETIC BASIS OF INTELLIGENCE
heritability coefficients for the working
memory tasks ranged from .43 to .48, de¬
pending on the task.57
In the United States a twin study was con¬
ducted utilizing records maintained by the
US Department of Veterans’ Affairs (VA),
which is responsible for ongoing studies of
the health of military veterans.58 Just under
350 pairs of male twins, ranging in age from
forty-one to fifty-eight, were given the read¬
ing span measure of working memory (see
Chapter 6}. The estimate of h 2 based on
the difference in correlations between MZ
and DZ twins was .58. This was reduced
slightly by more sophisticated modeling, but
it was clear that the estimate should be in the
.5-6 range. Further analyses indicated that
the correlation between the reading span
measure and measures of reading skill were
mediated genetically.
These four studies, done with partici¬
pants of different ages and conducted by
different laboratories in different countries,
have produced consistent results. Work¬
ing memory, one of the key information¬
processing underpinnings of intelligence,
is subject to substantial genetic influence.
However, the influence seems to be some¬
what smaller than the genetic influence on
measures of g.
Establishing the genetic basis for speed
of cognitive processing is a bit compli¬
cated. Speed of cognitive processing has
been shown to be a component of intel¬
ligence apart from the storage and atten-
tional control components of working mem¬
ory (Chapter 6). However, in above-average
young adults (usually college students) the
contribution of processing speed to conven¬
tional psychometric measures of intelligence
is considerably smaller than the contribution
of the working memory functions. By con¬
trast, the decline in measures of psychome¬
tric intelligence from middle age onward is
very largely associated with declines in speed
of processing. This leads us to suspect that
studies of college students may not tell us all
that we need to know about the importance
37 Ando, Ono, & Wright, 2001.
38 Kremen et al., 2007.
239
of speed of processing in the population at
large.
There are many ways to measure speed
of cognitive processing. Most conven¬
tional intelligence tests use pencil-and-paper
methods, where the dependent measure is
how many easy tasks can be accomplished in
a fixed time period (e.g., simple additions).
Researchers, especially those who have dealt
with information-processing measures in
contexts other than studies of individual dif¬
ference, prefer the tighter control afforded
by computer-controlled tasks, such as the
choice reaction time (CRT) and inspection
time (IT) measures. Within each of these
broad paradigms, and especially within the
information-processing paradigm, there are
many variations in procedure.
In spite of these difficulties, studies of
the genetic basis of speed of cognitive pro¬
cessing have been remarkably successful. A
meta-analysis of several studies concluded
that heritability was about .18 for easy timed
tasks (which do not have high correlations
with test scores) and .52 for hard tasks.59
Looking at the details of a few of the studies
is informative.
In the study of Dutch twins referred to
earlier, the investigators found that, depend¬
ing on the particular speed-of-processing
task, heritability coefficients ranged in the
.4-. 5 range.40 Moving to adolescents, investi¬
gators in the Colorado adoption study (panel
8.4) studied the genetics of speed of process¬
ing in sixteen-year-olds, using pencil-and-
paper tasks. This study found a speed of
processing heritability of .48. 41 Literally on
the other side of the world, the estimated
heritability of inspection time was .80 in the
Australian study of sixteen-year-old twins.42
This is the highest reported value of heri¬
tability that I know of.
One of the studies associated with the
Dutch Twin Registry examined heritability
in two cohorts, one aged twenty to thirty
at the time of the examination and another
39 Beaujean, 2005. See also Jensen, 2006, p. 130 ff., for
some brief notes concerning other studies.
40 Polderman et al., 2006.
41 Alarcon et al., 1999.
42 Luciano, Smith, et al., 2001.
240
HUMAN INTELLIGENCE
Figure 8.11. Heritability estimates for three academic skills,
based on data from the TEDS study (Chapter II). NC refers
to teachers' ratings of student progress on the United
Kingdom national curriculum in reading or mathematics.
aged forty to fifty-five. The heritability
estimate for the inspection time measure
was .46, and did not differ across cohorts.45
The Swedish study of twins raised apart44
contained participants from fifty to eighty
years old. The study included pencil-
and-paper measures of perceptual speed.
Because this was a repeated-measures study
it was possible to estimate the extent of
heritability and environmental influences on
both overall perceptual speed (as evaluated
by pencil-and-paper testing) and changes
in perceptual speed. The heritability coef¬
ficient was around .8 for overall speed
(depending slightly on which of two tests
was used to measure speed), but the major
influence on change was the nonshared envi¬
ronmental difference.
8.3.5. Q^G Analyses of Academic Skills
Conventional intelligence tests evaluate
important cognitive traits, but any reason¬
able theory of intelligence has to include
cognitive traits that are not evaluated by the
tests, and quite possibly cannot be evaluated
in the context of the time-limited “Drop
in from the Sky" testing session. When dis¬
cussing the genetics of intelligence it makes
sense to ask what contribution genetic inher¬
itance makes to cognitive skills that are
important in the world, but lie outside the
conventional testing realm. Among the most
important of these are skills in reading and
elementary mathematics, because of their
central role in school and the workplace.
The largest study of genetics and aca¬
demic achievement today, and probably the
largest that will be conducted for some
time, is the Twins Early Development Study
(TEDS), a United Kingdom study of twins
involving 12,000 participating families.45 The
study is described in panel 8.8. Testing was
carried out in infancy, where the evalua¬
tion emphasized language development, and
again at seven, nine, and ten years of age.
At each year age-appropriate evaluations
were made of children’s progress in English-
language studies, mathematics, and science.
These evaluations were supplemented by
teachers' ratings of student progress.
Figure 8.11 shows the estimated heritabil¬
ity for two classes of variables: teacher rat¬
ings and tests of children’s mathematics,
reading, and science levels at three different
ages. The estimates range from a minimum
of .4 to a maximum of .7. In all cases the
heritability estimate exceeded the estimate
for percentages of variance associated with
either shared or nonshared environments,
43 Posthum a ( de Geuss, & Boomsma, 2001.
44 Finkel & Pedersen, 2004; Reynolds et al., 2005. 45 Kovas et al., 2007.
THE GENETIC BASIS OF INTELLIGENCE
241
Panel 8.8. The Twins Early
Development Study (TEDS)
TEDS was a study of over 12,000 families,
recruited from some 25,000 families who
had twin births in the 1994-96 period,
according to the United Kingdom (UK)
National Health database. The children
were tested when they were seven, nine,
and ten years of age as, according to the
UK’s curriculum plan, these represent
critical ages in cognitive development.
In-person testing was generally not
possible. Therefore, the investigators
relied on two sources of data: teachers’
ratings of how well the students were
doing in Reading, Mathematics, and Sci¬
ence, and some cleverly designed "dis¬
tance tests" that could be administered
(with parental cooperation) over the tele¬
phone or, for older children, over the
World Wide Web. Verification studies
involving in-person testing were con¬
ducted in order to check on the accu¬
racy of the distance tests. Teacher rat¬
ings and the test results told basically the
same story. Heritability was the largest
single factor determining variation in
all skills. Nonshared (basically withm-
family) environment was the second-
largest factor.
The conclusions of the TEDS study
that relate to our understanding of intel¬
ligence are presented in the main text.
Here I will consider two further aspects
of this large, important study; how widely
its results can be generalized, and what
these results mean for education, espe¬
cially special education programs for chil¬
dren who are not doing well.
TEDS found that in a broad sense
genetic heritability was the single largest
influence on test scores. There was no
evidence that very poor performance, in
the bottom 5% or 15% of the popula¬
tion, was due to some special condition
(e.g., a genetic condition that impacted
ability to acquire language skills). Putting
this somewhat colloquially, according to
the TEDS results many of the children
who, in the US, would qualify for spe¬
cial education in reading and mathemat¬
ics aren ’ t really special; they are j ust at the
lower end of the distribution of reading
and mathematics skills. Against this, we
have to remember that there are specific
genetic bases for a number of moderate
mental disabilities, including reading dis¬
abilities. By lumping together all children
below a certain ability level, the TEDS
investigators may have developed a pic¬
ture of disability that is accurate at a large
scale, but that overlooks some important,
albeit uncommon, specific disabilities.
Environmental effects upon the means
of affected groups are well established.
For instance (as is documented in the
next chapter), intensive preschool edu¬
cational programs can improve the aca¬
demic readiness of children from very
low SES groups. This indicates that bet¬
ter school environments can help, a point
that has been amply demonstrated in a
variety of settings.* Whether these pro¬
grams will do anything to diminish the
variation in performance between chil¬
dren is another matter.
The fact that genetic influences are
so strong suggests that genetic analysis
could serve as an early warning signal for
identifying children who might have diffi¬
culty acquiring language and mathemat¬
ical skills. The genetic analysis could be
as simple as identifying a child whose rel¬
atives have had trouble with academics
in the past, or it could be as complex as
actually identifying the child’s genome.
We are not quite ready for the latter sort
of testing. Leaving aside the costs of such
a program, which are dropping rapidly,
we do not know, yet, just what genes
and alleles to look for! In addition, using
genetic indicators as guides for educa¬
tional decisions raises social policy issues
that are quite beyond the scope of this
book.
The TEDS analyses of children at the
low end of mathematics, reading, and
(continued)
242
HUMAN INTELLIGENCE
Panel 8.8 ( continued )
science competency indicate that, in the
vast majority of cases, these children rep¬
resent the low end of normalcy rather
than specific genetic or environmental
defects. This has no implications at all
for policy. Setting the level at which
schoolchildren are considered for special
education, in any field, has to be deter¬
mined by two things: the level of com¬
petency the child requires in order to
function in the society, and the money
available to enroll students in special edu¬
cation programs.
Why a particular child falls into the
defined class is of interest only if it serves
as a guide to treatment, to tell the teacher
how to fix the problem. The finding that
the special class is actually the low end of
the normal distribution is encouraging,
because it indicates that the special edu¬
cation class will be reasonably homoge¬
neous, so that uniform teaching methods
may apply widely. What those methods
should be is a topic beyond the current
discussion.
* Nisbett, 2009, provides an extensive list of exam¬
ples and discusses the difficulty of implemen¬
tation.
and heritability often exceeded total envi¬
ronmental influences.
Genetic correlations were high, indicat¬
ing that substantially the same genetic influ¬
ences were being expressed in all topics.
This is consistent with QBG analyses of
psychometric tests [as discussed earlier);
genetic influences appear to express them¬
selves more on g than on specific cogni¬
tive skills. The environmental influences
appeared to be unique to each of the three
skills - reading, math, and science. Environ¬
mental effects appeared to control devia¬
tions from a genetically related stable path,
either upward or downward, rather than
individual differences in the average per¬
formance of a child, across time. As has
been the case with studies of genetic influ¬
ence on intelligence test scores, the envi¬
ronmental contribution was largely due to
nonshared environmental differences, envi¬
ronmental differences between twins or sib¬
lings, rather than across families.
Educators [and parents) have long
debated whether particularly poor school
performance is an indication that a child
has a specific genetic anomaly or whether
children toward the bottom end of the
performance scale have general educational
deficits. To shed light on this question the
TEDS researchers asked whether the mod¬
els that applied to all children in the study
also applied to those in the lowest fifth
or fifteenth percentiles. In the American
context, low-performing children are often
assigned to “special education” classes. The
reduced samples still contained several hun¬
dred cases. In general, the same models that
applied to the entire sample applied to the
reduced samples.
This suggests that poor academic perfor¬
mance is generally the result of an overall
cognitive deficit, rather than a specific prob¬
lem. It is true that some cases of impaired
learning skills have been identified with cer¬
tain genomes.46 However, the known spe¬
cific disabilities account for only a small frac¬
tion of cases of poor reading or mathematics
performance.
8.3.6. Summary Comments on
Quantitative Behavior Genetics
Anyone familiar with the “now you find
it, now you don’t” phenomena that plague
the social sciences has to be struck by the
46 Willcut et al., 2004. An example of such specific
genetic anomalies is the FOXP2 gene, which was
associated with very poor linguistic performance in
a pedigree study of a single British family (Lai et al.,
2001). The relevant allele was not found in any of
the 270 cases of poor language performance in the
TEDS study (Kovas et al., 2007, p. 108).
THE GENETIC BASIS OF INTELLIGENCE
consistency of the behavior genetic findings.
To review briefly:
1. Additive genetic heritability accounts
for 40-80% of the variance in virtually
all cognitive traits.
2. Somewhere around half of the variance
is common to all traits. This strongly
suggests that there are generalist genes
that influence brain structures underly¬
ing virtually all cognitive performance.
3. Environmental effects (including the
prenatal environment) are strongest in
early childhood. They diminish there¬
after.
4. Childhood environmental effects are
primarily due to within-family differ¬
ences - the way a child interacts with
his or her familial environment - rather
than being due to between-family envi¬
ronmental differences. Nonshared envi¬
ronmental influences are also the largest
environmental influences in adults, but
this now refers to life experiences
that differentiate twins and siblings,
rather than (in most cases) to family
environments.
5. While genetic influences are strong in
establishing the trajectory of a person's
cognitive development and the later
reduction of cognitive ability in old age,
variations from that trajectory are influ¬
enced by environmental factors.
There are a few caveats that must be kept
in mind. I will present them in inverse order
of their importance, from least serious to
most serious.
Twin studies loom large in the QBG
study of intelligence. Heritability estimates
from such studies are heavily influenced
(and in some cases dictated) by the contrast
between DZ and MZ twins. The rough and
ready method of estimating the heritability
of trait X as 2 (rMz(X, X') - rDX(X, X')) is
a good example. This contrast is based on
the assumption that MZ twins have identi¬
cal geneotypes, which they do, and that DZ
twins have half of the permissible human
variation: rMz(G, G) = 1 and roz(G, G7) =
V2. The assumption about MZ twins is unar¬
guable. The assumption about DZ twins
243
depends on the further assumption that the
parental genetic correlations are zero; there
is no tendency for males and females to
select mates of similar genetic background.
This may be true in fruit flies and fish, but it
is not in humans. The correlation between
cognitive test scores for spouses generally
runs in the .2-3 range. This tendency is
called assortative mating.
As yet, we do not have a good idea of
the extent to which assortative mating has
to be considered. The topic is a difficult one
to study, because the analysis requires test
scores from both parents and children. Nev¬
ertheless, some studies are being done, and
hopefully more will occur in the future.47
Most of the analyses used to deter¬
mine heritability do not allow for gene-
environment interactions or correlations.
The failure to allow for interactions is a
major deficiency of present QBG ap¬
proaches. Why is it a problem, and why
haven't QBG analyses allowed for it?
We know gene-environment interactions
exist, for we can point to some. Many genet¬
ically influenced traits, such as alcoholism,
involve a potential for a pathology that is
released by environmental agents. As we get
a better understanding of gene-environment
interactions they are likely to play a larger
role in our understanding of the inheritance
of intelligence. Gene by environment cor¬
relations may also be very important. Sev¬
eral studies have reported higher heritability
coefficients in middle and upper SES pop¬
ulations than for lower SES populations.48
This is in itself an interaction in the statisti¬
cal sense. But why would this occur? Several
mechanisms have been postulated.
One argument is that harsh social envi¬
ronments will restrict the expression of
genetic potential, and that therefore genetic
variation can express itself only in an envi¬
ronment that nurtures cognition 49 There is
an analogy to weight: in a famine-stricken
town everyone is emaciated; when food is
47 See Colom, Aluja-Fabregat, & Garcia-Lopez, 2002;
Reynolds, Baker, & Pederson, 2000; van Leeuwen,
Van den Berg, & Boomsma, 2008.
48 Harden, Turkheimer, & Loehlin, 2007; Rowe, 2004;
Turkheimer et al., 2003.
49 Bronfenbrenner & Ceci, 1994.
HUMAN INTELLIGENCE
244
available genetic tendencies toward obesity
can express themselves. Substitute “getting
something out of education” for weight and
“bad schools” for famine and you have the
idea.
A second argument is that society con¬
tains a number of positive feedback mecha¬
nisms that serve to increase variation, espe¬
cially in the high range of the scale. Jim
Flynn, a professor of political science at
the University of Otago, New Zealand has
been a particularly strong advocate of this
hypothesis.50 One of his frequently used
analogies is to basketball. Initially a tall,
strong young person may be singled out for
special playing and coaching opportunities.
The better the player, the more coaching the
player gets, thus increasing the variance in
skill between the initially mediocre and ini¬
tially somewhat better players. Applied to
cognitive skills, this suggests that if a school
system practices “streaming” talented and
untalented students onto different tracks,
with different qualities of instruction, one
can anticipate a rich-get-richer, poor-get-
poorer phenomenon, in both basketball and
intelligence. Initial, genetically produced
talent is nurtured by the environment.
Both these explanations of the observed
changes in hz with environment rest on
interactions between initial talent and
opportunities to develop talent. There is a
third argument that is equally reasonable
but has an entirely different basis. The issue
may not be that higher SES environments
permit the expression of greater genetic
effects; it may be that lower SES environ¬
ments have greater environmental ranges in
variables that are relevant to intelligence. To
put things graphically, the cognitive benefits
of sending a child to a good, average middle
class preschool, or even of not sending the
child to preschool at all, but spending a lot of
time interacting with the child at home, may
not be very far behind the benefits of send¬
ing a child to an exceptionally expensive
preschool. The differences between keeping
children in a chaotic atmosphere, with very
little encouragement to explore or express
themselves, compared to sending the same
50 Dickens & Flynn, 2001; Flynn, 2007.
children to a modestly funded preschool
three or four days a week may be profound.
This sort of effect is not a gene-environment
interaction. The effect is due to greater envi¬
ronmental variation in the lower SES envi¬
ronment compared to the higher SES
environment.
I find these arguments for gene-
environment interactions and correlations
compelling. However, I see very little hope
of untangling the effects by the use of
structural equation models that make gross
assumptions about how the environment
covaries between people of different degrees
of genetic relatedness. We need direct mea¬
sures of the environmental variables that
influence cognition. Without a theory of
environmental action it is hard to know
where to begin.
And that is the rub. In an article defend¬
ing much of the research on twin stud¬
ies and, by implication, most other QBG
studies of intelligence, Bouchard 51 pointed
out that most of the claims that addi¬
tive heritability coefficients are due to hid¬
den gene-environment interactive effects are
arguments about what might be the case.
The arguments are sometimes accompanied
by analogies to gene-environment interac¬
tions in plants, but seldom if ever by an
example involving human intelligence. In
this respect, Flynn’s apocryphal basketball
example is interesting, but an actual exam¬
ple involving human intelligence would
have been more convincing.
Bouchard made two arguments to
counter the interactionists. Both have to
do with the strategy one follows in scien¬
tific research. First, he observed that many
of the arguments appealed to what ought
to have been measured in various studies.
My own concerns, as just expressed, are of
this nature. Any result can be explained by
appeal to unmeasured variables.52 Second,
scientists generally accept a principle known
as Occam's razor. Given two equally accu¬
rate explanations of the same phenomenon,
one should always prefer the simpler one.
51 Bouchard, 1997.
52 1 wish I could take credit for this statement, but I
cannot. I heard Bouchard say it in a public meeting.
THE GENETIC BASIS OF INTELLIGENCE
QBG studies have shown that the assump¬
tion of substantial additive genetic variance
can account for a great deal of data on
the covariation of intelligence across peo¬
ple of different degrees of genetic related¬
ness. Therefore; the simple additive model
deserves to be number one in a competi¬
tion between theories. Until someone comes
along with data to back up the more com¬
plicated interaction models.
QBG tells us how much heritability there
is for intelligence within a given population.
The only thing it says about the mechanism
of inheritance is an appeal to the abstract
Mendelian concept of a gene. To go further
we have to ask what a gene is and how it
influences the development of the brain.
8.4. The Molecular Genetics of
Intelligence
Mendel established the logical basis of
genetic inheritance; but knew nothing of
its physical mechanism. That is the topic of
molecular genetics, a field that started at the
beginning of the twentieth century with
the discovery that the chromosomes were
the bearers of the genetic material. Dis¬
covery after discovery followed (along with
several Nobel Prizes); the key one being
James Watson and Francis Crick’s discovery
of the structure of the genetic material;
deoxyribose nucleic acid (DNA); in 1953. 53
Genetic inheritance turns out to be a
very complicated process. Fortunately a
simplified model of the process is sufficient
for our purposes.54
The material in the chromosome is a
DNA molecule; which has the structure of a
double helix, two strands of material twisted
around each other. The strands are made
up of four structures called bases. The four
bases are adenine (A), thymine (T); guanine
(G); and cytosine (C). The strands of the
helix are bound together because the bases
bind to each other, forming units called base
pairs. The binding is unique. An A always
53 Watson & Crick, 1953.
54 Watson (2003) presents a well- written history and
discussion of the twentieth-century discoveries.
245
pairs with a T, and a C with a G. Logi¬
cally the DNA molecule can be thought of
as a sequence disregarding its helical struc¬
ture. The sequence constitutes a person's
genome. This is why the term "sequencing
the genome" is sometimes used to refer to
the process of identifying a genome.
The bases form triplets, sequences of
three bases. As there are four possible bases,
there are sixty-four possible triplets. These
are called codons. Each of the sixty-four pos¬
sible codons either marks the starting and
stopping points of the codon sequence that
makes up a gene, or specifies the creation
of one of twenty amino acids, which are
the building blocks of enzymes and pro¬
teins out of which cells, and hence life, are
constructed. Therefore, a protein-encoding
gene is, physically, the sequence of codons
between a “starting” and “stopping” codon.
The sequence of bases within this inter¬
val provide the program for cellular mecha¬
nisms that initiate the construction of one or
more proteins, just as the magnetic charges
on a digital video disk (DVD) provide the
program that initiates the display of pictures
and playing of music on a modern television
set. Since Watson and Crick's discovery, one
of the major tasks of behavior genetics has
been to find out how this program is read to
produce proteins, then cells, and eventually
an organism.
Variations in the sequence of bases within
a gene define alleles, and hence variations
in the program for building the organism.
This provides the variation that evolution
requires.
Figure 8.12 shows how genes are passed
on from one generation to the next. The
figure illustrates meiosis, the process of con¬
structing a chromosome carried by a gamete
(sperm cell in males, ovum in females).
Recall that (for autosomes) there are two
chromosomes. The two chromosomes link
at a connecting point. A new chromosome
is formed, containing a segment from each
of the parental chromosomes. Therefore,
base pairs that are located close to each
other on a chromosome tend to be inherited
together. In fertilization the chromosomes
in the sperm and ovum unite, so that the
246
HUMAN INTELLIGENCE
Figure 8.12. The reshuffling of genetic material in meiosis. Each
chromosome in a chromosome pair can be thought of as a sequence
of base pairs. In meiosis the two chromosomes link together at some
midpoint, and then a new chromosome is constructed from one
segment of each of the parental chromosomes. Therefore, base pairs
that are close to each other on a chromosome will tend to be
inherited together.
chromosome pair in the offspring contains
one chromosome from each parent.
The process is somewhat different in the
sex chromosomes. In females, who have two
X chromosomes, the shuffling takes place
as in the autosomes. In males, who have
an X and Y chromosome pair, either the
X or Y chromosome is carried over into a
sperm cell. Accordingly, female zygotes con¬
tain two X chromosomes, one from each
parent, but male zygotes contain a mater¬
nal X and a paternal Y chromosome. The Y
chromosome represents a direct transfer of
genetic material along the patrilineal line.
While a male’s X chromosome is always
inherited from the mother, that chromo¬
some might have come from either mater¬
nal grandparent, so an analogous continuous
transfer of genes through the female line is
not possible.55
The white and black dots in Figure 8.12
indicate base sequences. The black dots indi¬
cate sequences that make up the protein-
55 There is a way to trace female lineage. Mitochon¬
dria are organelles within a cell, but outside the
nucleus. They contain their own DNA, mtDNA, for
a total of thirty-seven genes, plus segments of non-
coding DNA. MtDNA is inherited from the mother,
so analysis of mtDNA provides a genetic record of
female lineage.
encoding genes. The white dots indicate
base sequences outside of the genes. The
totality of the base sequences, across all
chromosomes, constitutes the genome.
About 90% of the genome consists of
sequences of bases that lie outside of
the protein-encoding genes. There are also
sequences of bases within the protein¬
encoding genes that do not code for proteins.
At one time these sequences were thought
to be inert and were referred to as "junk
DNA." We now know that at least some of
the sequences are active, because they regu¬
late the expression of the protein-encoding
genes. In addition, some genes regulate the
expression of other genes.
There are many statements like "We share
50% of our genes with each of our par¬
ents." These statements are a bit mislead¬
ing. The 50% figure refers to variation in
the sequence of base pairs that can differ
across human beings, and still produce a
person. That is a small percentage of the
total genome; roughly 99.9% of the genome
is identical across humans, and we share 96%
of the genome with our evolutionary cousin,
the chimpanzee.
Nevertheless, there are differences. A
difference that occurs at a base pair in
THE GENETIC BASIS OF INTELLIGENCE
247
Panel 8.9. A Few Stray Facts About
the Genome
Here are a few facts about the genome.
1 . There are approximately three bil¬
lion (3,000,000,000) base pairs in
the human genome. However, the
number of protein-encoding genes
is much smaller. The current (2010)
estimate is between 25,000 and
30,000. This estimate is well below
the estimates of too, 000 or more that
were common as late as the year 2000.
2. Each gene specifies an average of
three proteins. Different proteins
may be expressed in different parts
of the body. It has been estimated
that about one-third of human genes
have some expression in the brain.
3. A mutation occurs when a new
allele arises. Some mutations arise
due to errors in the process of
DNA replication. Mutations can also
be triggered by environmental haz¬
ards. These include nuclear radia¬
tion, ultraviolet light, and exposure
to certain chemicals. Discouragingly,
some of the hazardous chemicals are
used in industrial processes. Finally,
some mutations appear to occur “by
chance,” which simply means that
we do not know what the causative
process is.
4. Mutations are probably more fre¬
quent than we realize, for a great
deal of the genome’s programming
specifies basic processes in the cell.
In such cases pregnancy either does
not begin or is terminated without
detection. In other cases termination
occurs later, with a detectable preg¬
nancy, or a stillbirth occurs. Some
mutations are viable, and can even
produce superior individuals in some
environments. That is how evolution
works!
5. Some genes regulate the expression
of other genes in the body. In addi¬
tion, some of the (formerly believed
to be) junk DNA lying outside
the boundaries of protein-encoding
genes serve as regulators of genetic
expression. Microenvironmental fac¬
tors within the cell can also control
genetic expression.
6. A genetic influence may not be
expressed for some time after birth,
or in some cases may be suppressed
throughout the individual’s life.
at least 1% of the population is called a
single nucleotide polymorphism (SNP or
“snip”). There are about ten million SNPs
in humans. Blocks of SNPs that are close
together on the same chromosome, and so
tend to be inherited together, are called hap-
loids. The SNPs themselves may be in a
protein-encoding gene, in a region between
genes, or may be part of the nonencoding
portion of a gene. One of the major steps
in research on genetic differences involves
locating haploids that differ between two or
more populations of interest, such as indi¬
viduals who do or do not have a particular
form of mental disability.
Panel 8.9 presents a few more facts about
the genome. The next two sections discuss
some of the many genetic disorders of
cognition.
8.4.1. Genetic Pathologies of Intelligence
A person is considered to be mentally dis¬
abled if his or her intelligence, as measured
by standard tests, falls below an IQ of 70.
This is obviously an arbitrary standard, for
a great many people who are considered to
have a mental disability have IQ scores in
the 70-80 range. From the behavior genet¬
ics viewpoint, the mentally disabled fall into
three distinct groups.
One group consists of people who have
a cognitive deficiency with a known genetic
cause. The mental deficiency is often only
248
HUMAN INTELLIGENCE
part of the syndrome, which may often
include other organic problems. There are
approximately 300 such syndromes known
at present, and no doubt more will be dis¬
covered. In some cases there is a general
mental deficiency; in others deficiencies may
be unevenly distributed, although some gen¬
eral deficiency is usually found. The genes
that have been identified as causes for these
deficiencies are found on several different
chromosomes. However, the rate of occur¬
rence on the X chromosome is higher than
would be expected by chance.56
Most of the genes associated with severe
mental retardation are recessive. Accord¬
ingly, a man can suffer ill effects if he inher¬
its a single copy of the allele associated with
the pathology. Women, with two X chro¬
mosomes, have to inherit two copies of the
allele in order to be affected.
The following examples of known genetic
disorders have been chosen to illustrate the
variety of syndromes that can occur. One
of the examples is expressed in infancy, one
in infancy and childhood, another in mid¬
dle age, and the last is usually not expressed
until old age. In the first three examples the
genetic anomaly is both sufficient and neces¬
sary for the disorder to occur. In the fourth
example the genetic anomaly increases the
risk of the cognitive disorder, but is neither
a sufficient nor a necessary condition for the
disorder.
Phenylketonuria (PKJJ). This is one of
the best understood of the genetic mental
deficiencies. PKU is due to a mutation in
the PAH gene located on chromosome 12.
PAH codes for synthesis of phenylalanine
hydroxylase, an enzyme that is important
in the metabolism of phenylalanine, a sub¬
stance that is found in certain foods, includ¬
ing artificial sweeteners. Untreated PKU can
result in severe mental retardation. Fortu¬
nately the condition leading to PKU can be
detected at birth, by an inexpensive blood
test. Affected individuals have to maintain a
restricted diet for a number of years. There
are some indications that older children and
adults who cease to follow the diet may be
56 Inlow & Restifo, 2004.
at risk for mild mental retardation, or at least
a lower tested intelligence level than would
be expected otherwise. In the extreme case
a pregnant woman who is a carrier of the
mutation may affect her child, not necessar¬
ily by passing on the gene, but by the harm¬
ful effects of a buildup of phenylalanine in
her body during pregnancy.
Estimates of the frequency of PKU range
from 1 per 13,500 to 1 per 19,000 births,
varying across ethnic groups. Screening and
treatment for PKU has been widespread
since the 1960s, so the clinical form of the
condition is now fairly rare in the industri¬
ally developed countries. This makes PKU a
prototypical example of a gene-environment
interaction. It is a disaster for children born
without access to modern health care, but
a treatable problem for children who have
access to health care.
Fragile X syndrome. People suffering from
fragile X syndrome display impulsiveness,
difficulties in concentration, and, possibly
concomitantly, some degree of mental retar¬
dation. The condition affects 1 in 4,000 men
and 1 in 8,000 women. It is due to a muta¬
tion in the FRMi gene on the X chromo¬
some. This gene is involved in production of
a protein that is important in, among other
things, nerve synthesis.
Molecular geneticists have been able to go
inside the gene, pointing to the manner of
mutation that influences the gene's expres¬
sion. In normal individuals the CGG codon
occurs in a block of from ten to forty rep¬
etitions within the FRMi gene. In affected
individuals the block may contain from 200
to 1,000 repetitions. The repetitions block
the gene's normal production of the protein.
Huntington's disease. Huntington’s dis¬
ease was discussed earlier as an illustra¬
tion of a pathological gene that does not
express itself until after a person reaches
or passes reproductive age. See the discus¬
sion of Woody Guthrie and his family, pre¬
sented in panel 8.1. The syndrome is caused
by a dominant allele on chromosome 4, so
affected individuals need have only one copy
of the gene in order to express symptoms.
In general, if a pathological allele is
dominant it quickly disappears from the
THE GENETIC BASIS OF INTELLIGENCE
population. Huntington’s disease represents
an insidious situation. The pathology does
not manifest itself until after the affected
individual is well into his or her reproduc¬
tive period. Therefore, the disease gene can
be passed on to the next generation before
the illness appears.
As is the case with fragile X syndrome,
the pathological allele has an abnormal num¬
ber of repeats of a codon. Increasing num¬
bers of repeats are associated with earlier
onset of the syndrome. There is also evi¬
dence that the length of the repeat increases
over generations, so if a child of a person
with Huntington’s disease carries the allele
he or she is at risk for developing symptoms
earlier than the parent did.
Alzheimer's disease. 57 Alzheimer’s disease
is a common and feared disease of the
elderly. As of 2008, estimates of the num¬
ber of people suffering from Alzheimer’s
dementia in the United States ranged from
4.5 to 7 million, out of a total population
of just over 300 million - about 2% of the
population. As the age distribution shifts
toward higher percentages of elderly peo¬
ple, the incidence of Alzheimer’s dementia
will increase. If present incidence rates con¬
tinue there will be eleven to sixteen million
cases of the disease in the United States by
2050. Progress toward effective treatment or
prevention has been disappointingly slow.
Research on this economically and socially
devastating disease is being given very high
priority.
The disease comes in two forms. Early
onset Alzheimer’s is defined as Alzheimer’s
disease that manifests itself prior to age 65.
Cases have occurred as early as the mid for¬
ties. The more common variety, late onset
Alzheimer’s disease, appears after age sixty-
five. The initial symptoms are failures of
short-term memory and attention, progress¬
ing to profound loss of long-term memory,
including loss of the ability to recognize
spouses and relatives known for fifty years or
57 Information on symptoms, causation, and incidence
downloaded from the Alzheimer’s Association web¬
site and the National Institute of Aging website,
March 2008.
249
more, loss of speech, and eventually death.
Because intensive care may be required for
a period of years the financial and emotional
impact on caregivers can be very high.
The proximal cause of the memory loss is
widespread deterioration of the nerve cells
in the brain, beginning with the frontal cor¬
tex and the hippocampus. This is not sur¬
prising, as these structures are important for
the development of new memories. Neural
deterioration is accompanied by the forma¬
tion of beta amaloid protein plaques in the
brain.
Three genes have been identified as in¬
creasing the risk of early onset Alzheimer’s,
but they are not sufficient to account for all
cases of the disease.
Late onset Alzheimer’s has been associ¬
ated with several genes, among them the
Apolipoprotein E (APOE) gene on chromo¬
some 19. The gene has three alleles, APOE2,
APOE3, and APOE4. About half the pop¬
ulation carries the APOE2 allele. The high-
risk form of the allele, APOE4, is relatively
common (15% of the European-derived pop¬
ulation in the United States). About 50%
of diagnosed Alzheimer’s patients carry the
APOE4 form, but less than half of the peo¬
ple who carry APOE4 express the disease.
For those who are affected, the age of onset
is related to the genetic load. Patients with
genotype E4/E4 have a mean onset of sixty-
eight years of age, patients carrying one
E4 allele have a mean onset of seventy-five
years, and patients with no E4 alleles have a
mean onset of eighty-four years.
None of the genes linked with either type
of Alzheimer’s disease has a strong enough
association to be considered the sole, or even
a necessary, cause of the disease. Identify¬
ing a person’s genotype makes it possible
to assess the risk of developing the demen¬
tia, but we cannot say “for certain" that the
dementia will or will not occur.
A number of environmental factors
have been linked (somewhat tenuously) to
Alzheimer’s disease. These include severe
inflammations of the brain and head injury.
It has been suggested that industrial air
and water pollutants may increase the risk
of the disease. This has yet to be proven.
250
HUMAN INTELLIGENCE
Such a multiplicity of possible causes is
not surprising. The proximal mechanism of
deterioration is neural degeneration, pos¬
sibly due to the development of protein
plaques in the brain, degeneration of the
synapses, and other insults to the brain itself.
The deterioration can probably be initiated
by several environmental and genetic haz¬
ards, which would act as distal causes.
Alzheimer's disease represents a common
situation, in which both genetic and envi¬
ronmental factors increase the risk of expres¬
sion of a pathology, but there does not
appear to be any one factor that is either
sufficient or necessary for the expression.
8.4.2. Mental Pathologies Related
to Chromosomal Abnormalities
Several mental disabilities are caused by
anomalies in the duplication of chromo¬
somes, rather than by variations in genes.
Down’s syndrome: Down's syndrome is
characterized by poor physical develop¬
ment, mild to moderate mental retardation,
and facial features that include elliptically
shaped eyelids. It is fairly common, occur¬
ring in approximately 1 of every 800 births.
The syndrome is caused by the presence of
an extra chromosome 21. Down's syndrome
sufferers usually die at a fairly early age, as
the syndrome includes defects in the cardio¬
vascular system as well as cognitive prob¬
lems. Adult Down’s syndrome patients dis¬
play premature aging, including the physical
deteriorations associated with advanced age
and early Alzheimer's disease.
Relatives of a person with Down’s syn¬
drome are at risk of producing a Down’s
syndrome child. The risk is strongly related
to the mother's age at birth. The risk rises
from less than 1 in 1,000 births to women
under thirty years old to 1 in 100 for women
over forty.
Turner's syndrome.^ Turner's syndrome
is a condition that affects women born with a
single X chromosome. Turner’s patients dis¬
play poor spatial-perceptual reasoning, the
P and R components of the VPR model.
58 Ross, Zinn, & McAuley, 2000.
Verbal reasoning is unaffected. Affected
women may also display difficulties with
tasks that place high demands on the com¬
plex of working memory and attentional
control, the behaviors labeled “executive
functioning.” Physical signs may appear.
These include short stature, in some cases
markedly thickened necks, and late and
sometimes incomplete development of sex¬
ual characteristics. The women are also at
risk for cardiovascular problems. All of these
remarks have to be qualified by the fact of
individual variation.
Turner's syndrome cases vary greatly.
Some of these differences are tied to envi¬
ronmental variables. Girls who receive both
growth hormone and estrogen replacement
therapy may reach heights in the normal
range and may be free of the physical signs
associated with the syndrome. Support¬
ive within-family social environments are
important. Estrogen therapy appears to be
effective in countering deficiencies in exec¬
utive control functions, although it does not
counter the deficiency in spatial-perceptual
reasoning. This is interesting, because dif¬
ferent brain structures are involved in exec¬
utive control and spatial-visual reasoning.
Clinically, the result is encouraging because
a great many of our everyday spatial-
perceptual tasks can also be accomplished
using verbal strategies.
Turner's syndrome is, like PKU, an exam¬
ple of a gene-environment interaction. A
baby girl born today with Turner's syndrome
has a far better outlook than she would have
had one hundred years ago, providing that
she has been born into a social environment
that can provide the necessary support.
XYY syndrome. Some males are born with
two Y chromosomes. The genotype is asso¬
ciated with mild mental retardation. XYY
males are typically large and fairly robust.
During the 1960s it was suggested that XYY
males are overly aggressive, to the point of
exhibiting dangerous criminal behavior. The
evidence for this was that there is an ele¬
vated incidence of XYY men among crim¬
inal populations.59 However, it does not
59 Jarvik, Klodin, & Matsuyama, 1973.
THE GENETIC BASIS OF INTELLIGENCE
25*
follow from this observation that XYY men
are likely to become criminals. Studies of
XYY’s in prison populations can establish
the conditional probability of being XYY,
given that the individual is known to be a
(tall] criminal, Pr[XYY\crime) , but that is
not the same thing as establishing the prob¬
ability of being a criminal, given that the
individual is known to be an XYY. The lat¬
ter figure, Pr(crime\XYY) , is the statistic that
would have to be high in order to justify
proactive monitoring of XYY cases. This is
difficult to establish, as the condition occurs
only in approximately 1 in 1,000 male births.
Two small prospective studies have been
done in which XYY’s were identified and
then followed for some time. 60 In both stud¬
ies the incidence of criminality among XYY
men was elevated above the general popu¬
lation but not above the population of men
of comparable IQ. An analysis of the types
of crimes committed indicated that that ele¬
vated criminality may be mediated by low
intelligence levels, rather than heightened
aggression.
Klinefelter’s syndrome. Klinefelter’s cases
are men who have genotype XXY. Kline¬
felter’s syndrome cases are fairly large men,
although they are not unusually robust.
The commonest cognitive characteristic is
poor language development, along with
an elevated incidence of mild to moder¬
ate mental retardation. This is consistent
with an fMRI study that revealed decreased
hemispheric specialization during language
processing among Kleinfelter’s patients, as
compared to control groups.61 Sexual devel¬
opment is retarded. There is some indica¬
tion that hormone replacement therapy can
ameliorate the symptoms of this condition.62
Various types of mental retardation affect
slightly under 3% of the population. This
estimate excludes people with Alzheimer’s
disease, which is usually considered a prob¬
lem associated with old age rather than
a mental disability. The mentally disabled
60 Gotz, Johnstone, & Ratcliffe, 1999; Witkin et al.,
i976.
61 Van Rijn et al., 2008.
62 Hazlett, 2005.
impose a substantial burden on our social
support systems. Finding techniques for
ameliorating or effectively eliminating the
consequences of these genetic anomalies,
as we have with PKU, would have major
moral, social, and economic rewards.
8.4.3. Tt te Genetic Basis of Normal
Variability in Intelligence
The molecular genetic analysis of normal
variation in intelligence presents a very dif¬
ferent picture than the analysis of patholog¬
ical conditions. The first thing to say, and
to say loudly, is that there is no one gene,
or even a small number of genes, responsi¬
ble for normal variations in intelligence. If
there were one, we would have found it by
now, for the techniques that have been used
are quite adequate to identify any gene that
accounted for 30% or more of the variation
in intelligence within the normal range. No
such gene has been found, so we may be
pretty sure that it is not there.
We are clearly dealing with a poly¬
genic inheritance model; lots of genes have
their influence, but no one of them is
the key gene. One of the most prominent
researchers in the field, Robert Plomin, has
speculated that variation in any one gene
will control at most .5% of the variance in
intelligence, as expressed in test scores. 63
This may be a bit pessimistic, for researchers
have found at least two genes, the CHRM2
gene on chromosome 2 and the SNAP-25
gene on chromosome 20, where variations in
alleles may account for as much as 3-4% of
phenotypic variation.64 However, such find¬
ings have to be replicated. And here is why.
Population-wide studies of genetics assess
covariances between allele frequencies
and phenotypical variations simultaneously,
across many polymorphisms in the geno¬
type. Because very many polymorphisms are
being studied simultaneously, there is a sub¬
stantial chance of finding polymorphisms
that, by chance, happen to covary with
the phenotypical measure in the sample,
63 Plomin, Kennedy, & Craig, 2006.
64 Dick et al., 2007; Gosso et al., 2006.
252
HUMAN INTELLIGENCE
even though there is no covariation between
the allele frequencies and the phenotypical
measure in the population. Such a finding is
called a false positive.
The converse problem is false negatives.
Suppose that Plomin’s conjecture is right,
that the covariation between allele frequen¬
cies in any one gene and intelligence test
scores accounts for less than 1% of the total
variation in test scores. Unless the sample is
very large the chances of detecting a small
covariation are not good.
The false positive and false negative
effects play off against each other. Any
widespread screening study maximizes the
chances of finding false positives. When it
comes to replication, though, a small effect
can dip below the level of detectability in
the replication sample. This makes the busi¬
ness of gene locating both time-consuming
and expensive. That is the way it is.
8.4.4. Techniques for Identifying Genes
Associated with Normal Intelligence
Several techniques have been used to iden¬
tify genetic contributions to intelligence.
Some of the technologies are described in
panel 8.10. Here I look at the logic rather
than the mechanics of any one method.
In a “bottom up” technique a gene is iden¬
tified that is known to be associated with
some physiological function that might, rea¬
sonably, be linked to intelligence. A study
is then initiated in which people are given
some form of intelligence test and have
their genotypes determined. This was the
method used to identify the CHRM and
SNAP genes. Both are believed to be linked
to the development of an efficient neural
system.
Sometimes this method works out. But
sometimes it does not. For instance, we
know that brain volume is related to intel¬
ligence, and that the connection is very
largely genetic.65 Accordingly, it is not sur¬
prising that there was considerable excite¬
ment when it was found that variations
in the allele frequencies of two genes,
65 Posthuma, De Geus, & Boomsma, 2004.
MICROCEPHALIN (chromosome 8) and
ASPM (chromosome 1], both known to
be involved in pathological failures of the
development of brain size, exhibit evidence
of strong evolutionary pressures and, in addi¬
tion, have a worldwide distribution suggest¬
ing that the pressure occurred about the
times of the migration of Homo sapiens
out of Africa and, much later, the devel¬
opment of agriculture. 6n However subse¬
quent research showed that allele variations
in these genes were not reliably associated
with intelligence, within the normal range,
or, for that matter, with head size and brain
volume within the normal range.67
Failures to replicate or to show hypoth¬
esized correlations do not uncover bad sci¬
ence in the original studies. Given the sta¬
tistical issues involved, the only way for the
field to proceed is for scientists to publish
their findings, and then call for replication.
An alternative strategy for finding the
genetic basis of intelligence is to start from
the "top down," by identifying individuals
who differ in intelligence and then contrast¬
ing their genotypes. This is called genome¬
wide association. One such study compared
DNA from children who had been identified
as showing a very high level of mathemat¬
ical ability to DNA from a normal control
group. The two groups differed in the fre¬
quency of an allele of a gene associated with
the synthesis of insulin.68
Because many DNA sequences are being
screened at once, genome-wide association
is prone to picking up false positives. The
only feasible way to counteract this is to
replicate findings. And, alas, when this
particular study was replicated the allele-
intelligence correlation was not found.69
Linkage analysis is usually applied in con¬
junction with the study of related individu¬
als - the pedigree method applied to people
rather than to plants. Linkage analyses can
identify segments of DNA where polymor¬
phisms are correlated with some measure
66 Evans et al., 2005; Mekel-Bobrov et al., 2005.
67 Mekel-Bobrov et al., 2007; Rushton, Vernon, &
Bons, 2007; Woods et al., 2006.
68 Chorney et al., 1998.
69 Hill et al., 2002.
THE GENETIC BASIS OF INTELLIGENCE
253
Panel 8.10. Methods for Screening
for Genetic Effects
This panel describes designs used to
establish a correlation between variations
in alleles [polymorphisms] and variations
in a phenotypic trait, such as intelligence.
Differences between individual ge¬
nomes [polymorphisms ) occur when the
sequence of nucleotides differs at one or
more positions, as in the sequence A C G
T A A compared to A C C T A A, where
there is a difference between G and C in
the third position. As the text explains,
a variation such as this is called a single
nucleotide polymorphism (SNP, “snip”].
A SNP can occur inside or outside the
boundaries of protein-encoding genes.
Segments of DNA that contain one or
more SNPs are referred to as alleles of
that segment, a generalization of the idea
that an allele is a variant of a gene to the
idea that an allele is a variant of a SNP.
SNPs occur roughly every 100 to 300 base
positions along the genetic code.
A genetic marker is an identifiable seg¬
ment of the sequence of bases in the
genome. Genetic markers are used to
identify segments of the genome within
each chromosome. To take an oversim¬
plified example, suppose that two mark¬
ers, Mi and M2, have been identified, and
that in one person the sequence of DNA
between markers is Mi A C G T A A M2
and in the other the sequence is
A C C T A A. There is a SNP at the
third position. This means that there are
at least two alleles of the genetic sequence
between Mi and M2.
Participants in a study donate DNA
samples and are measured on the phe¬
notypic trait, perhaps by being given an
intelligence test. The DNA samples are
then broken up into segments between
genetic markers. The data is examined
for correlations between the presence of
alleles and the values of the phenotypic
trait. For instance, if one allele is statisti¬
cally associated with high IQ scores and
another with low scores, then a genetic
association with IQ has been located.
The technique works best if the par¬
ticipants differ in their genotypes in
some systematic way. Therefore link¬
age analyses are done using samples of
related individuals. For instance, regis¬
trants in the Dutch twin registry and
their relatives have participated in linkage
analyses.
Linkage analysis is highly useful for
tracing the genetic basis of discrete traits,
with restricted reaction ranges, such as
eye color. It can be applied to contin¬
uously measured traits, such as intelli¬
gence, but there may not be enough reli¬
ability in measuring the phenotypic trait
to make the technique precise. This is
especially true if, as is the case for cog¬
nitive measures, expression of the trait
is affected by environmental as well as
genetic factors. We know that this is
the case for intelligence, for MZ twins,
with identical genotypes, do not necessar¬
ily have the same test scores. Therefore,
another method, genome-wide analysis, is
used with continuous traits.
Genome-wide analysis depends on a
technology for the detection of genetic
expression by microarrays. A microarray
is a small chip that is divided into cells, or
“dots.” In the screening application each
dot contains a sequence of DNA located
around one of the values of a previously
identified SNP. Therefore, the segment of
DNA in each dot is essentially a fragment
of the allele of a SNP carried by one indi¬
vidual in the sample. Chips capable of
evaluating a million SNPs have become
available.
DNA samples are taken from individ¬
uals from one or more populations that
vary on the trait in question. The genetic
material from all individuals is pooled, to
create a solution of DNA and other mate¬
rial. (The process, which is rather com¬
plicated, will not be further described.)
(continued)
HUMAN INTELLIGENCE
254
Panel 8.10 ( continued )
The microarray chip is then immersed in
the DNA. Because complementary DNA
strands will bind to each other, DNA that
matches the DNA at a dot on the chip
will bind to that location. The amount
of binding at each spot is then com¬
pared across groups, to see if differences
in binding can be related to group mem¬
bership. A case of this sort is described in
the text.
of cognitive competence, usually intelli¬
gence test scores. (The technique would
work just as well with measures of work¬
ing memory, attention, or any other mea¬
surable trait, including such things as height
and eye color.) The reason for using related
individuals is that the occurrence of the
polymorphism can be traced over genera¬
tions and across different relationships, in
order to establish a correlation between the
polymorphism and the trait of interest. To
date, segments of DNA that covary with
intelligence test scores have been located
on chromosomes 2 and 6. 70 However, we
have to wait for replication to solidify these
results. Once this is done, further studies
can be initiated to locate the genes within
or near the segment, including non-protein-
encoding DNA regulator sequences, and to
find out what they do.
Progress toward identifying the genes
associated with normal variations in cogni¬
tion has been slow. In a review of progress
as of 2006, Robert Plomin, Kennedy, and
Craig pointed out that discoveries in the
field have followed a discouraging pattern.
Initially genes emerge as reasonable can¬
didates for the “intelligence gene” because
of their involvement with neural efficiency,
brain development, or some other key func¬
tion that ought to be related to intelligence.
Interesting associations are found at first, but
the associations do not appear on replica¬
tion. The fact that this pattern has appeared
so frequently is what led Plomin to specu¬
late that no gene, alone, accounts for more
than .5% of the variance in intelligence.71
70 Luciano et al., 2006; Posthuma et al., 2005.
71 Plomin, Kennedy, & Craig, 2006. Plomin reiterated
this theme in an address to the International Society
for Intelligence Research, Amsterdam, December
2007.
However, from quantitative genetic studies
we know that in the populations studied at
least half, and possibly more, of the variation
in intelligence is due to genetic variation. It
seems clear that we are dealing with a very
large number of genes, each of which has a
small effect. Smoking slingshots, not smok¬
ing gunsl
The search for the genes for intelligence
is not a hopeless project. Solid results from
quantitative behavior genetics show that the
target genes exist. The technology for asso¬
ciating genes with traits is getting better and
better. Ultimately, we will locate the genes
for intelligence and will understand their
modes of action. This will be a huge step for¬
ward in understanding the biological basis
of intelligence. Unless there is a spectacu¬
lar breakthrough in technology, the search
is going to take a lot longer than people
thought it would at the start of the twenty-
first century.
But be of (some) good cheer. The last fifty
years of the twentieth century and the first
decade of the twenty-first did see spectacu¬
lar breakthroughs in the technology dealing
with molecular genetics. Who knows what
is to come?
8.5. Reprise and Commentary
The facts are incontrovertible. Human intel¬
ligence is heavily influenced by genetics.
Specific genetic anomalies are clearly asso¬
ciated with identifiable mental abilities, and
on a statistical basis substantial heritabil-
ity coefficients have repeatedly been found
in studies throughout industrial and post¬
industrial societies. The exact value for the
heritability coefficient, h2, may vary some¬
what from study to study, but that is to
THE GENETIC BASIS OF INTELLIGENCE
255
be expected. It is never zero, and it is
never one.
Hopefully, future studies will shift from
trying to find a “true value" for this coef¬
ficient to determining what characteris¬
tics of a population determine the relative
importance of genetic and environmental
influences on intelligence. The heredity-
environment debate has too often been
posed as a contest. It is not. We need to
find mechanisms for both. An unacknowl¬
edged problem that has hampered the field
is that while we have an excellent theory of
genetic variation, we do not have a compa¬
rable theory of environmental variation. It is
easy to point to good and bad environments
for the development of cognitive power. It
is much harder to provide some sort of met¬
ric to say how good or how bad a particular
environment is.
The current frontier in behavior genetics
has shifted from trying to show that intelli¬
gence is inherited to trying to find the mech¬
anism of inheritance. A good deal of suc¬
cess has been obtained in tracing the genetic
causes of serious cognitive deficiencies, such
as Huntington’s disease and PKU. We have
had much less success in identifying the
genes associated with variations in normal
intelligence. Why? The answer is simple:
this is a harder problem. It may turn out that
the strategy of looking for gene-intelligence
correlations directly is the wrong approach.
An alternate approach would be to look
for the brain mechanisms that are associ¬
ated with intelligence, and then try to find
the genes that influence the development of
these mechanisms.
The lay public has displayed an ambigu¬
ous attitude toward findings on the genetics
of intelligence. I will revisit this topic in the
final chapter of the book, where I look at
social controversies more generally. Here I
make only a few brief remarks.
Public opinion has accepted an analogy
between physical defects and extreme men¬
tal deficiencies, especially when accompa¬
nied by physical stigmata. The lay person
believes that it is reasonable to look for a
physical cause for an extreme mental defi¬
ciency, just as it is sensible to look for
a physical cause for, say, high cholesterol,
which does have a genetic component. Such
research has received strong public, politi¬
cal, and financial support. There is consider¬
ably more ambiguity about efforts to iden¬
tify a genetic basis for intelligence, in the
normal range. There are several reasons for
this.
Modem Western society distrusts hered¬
itary elites. The Age of Kings is definitely
over. On the other hand, it is demonstra¬
bly the case that health, wealth, and oppor¬
tunity for education, the trappings of high
socioeconomic status, do pass on from one
generation to the next. In the nineteenth
century Galton concluded that this was
largely a reflection of genetically endowed
merit. Today that view is decidedly unpop¬
ular. Any suggestion that cognitive compe¬
tence is largely inherited is seen as deny¬
ing a popular legend - that you can be
what you want to be, if you work hard
enough.
The view that Behavioral Genetics denies
the importance of individual opportunity
and effort is frustrating to those who believe
in the importance of the partial genetic
inheritance of intelligence, because no com¬
petent behavioral geneticist argues that
mental competence is completely inher¬
ited. Behavior geneticists agree with educa¬
tors that socially important cognitive skills,
which are far more important than test
scores, can, within broad genetic limits, be
acquired by a combination of education,
experience, and effort. An argument for an
inherited component of intelligence is not
necessarily an argument against the value
of education and effort. This point is often
not appreciated by those who attack genetic
models of intelligence.
Discussion of the evidence for a genetic
basis for intelligence has been confounded
with discussions of differences in cogni¬
tive competence between racial and ethnic
groups. Again the facts are quite clear; in the
world today certain ethnic groups, notably
African-derived populations, are getting a
smaller share of the economic/social pie
than European-derived populations. This
has been going on for a long time. Claims of
256
HUMAN INTELLIGENCE
genetic influences on intelligence are seen by
some as tantamount to a sort of Darwinian
justification of present economic and social
inequalities.
In fact, these are separate issues. The
vast majority of current studies in behav¬
ioral genetics have no direct implications,
whatsoever, for issues surrounding racial
and ethnic differences in intelligence, sim¬
ply because they are studies of the varia¬
tion of genes and intelligence within rather
than across racial/ethnic groups, and gener¬
ally within just one ethnic group, European-
derived populations. The causes of within-
group variation are not necessarily the same
as the causes of between-group variation.
Unfortunately this has not stopped people
from speculating, on both sides of the issue.
This issue is discussed in some detail in
Chapter 11.
Now let us look at some environmental
influences on intelligence.
CHAPTER 9
Environmental Effects on Intelligence
More of your conversation would infect my
brain.
Shakespeare, C oriolanus, act 2,
scene 1
Shakespeare was right (again). Every expe¬
rience we have leaves an imprint on our
brains, and from it, on our minds. Clearly
physical experiences can change the brain.
It has been claimed that early species of
Homo got a leg up on the evolutionary lad¬
der when they began to eat fish.1 In modern
days, lecithin , a substance found in a num¬
ber of foods, including fish and eggs, has
been studied to see if it can enhance learn¬
ing. (The results are mixed.) We do not stop
there; we concern ourselves with the social
environment. If we do not believe that social
experiences can affect the brains of children,
why do we have decency ratings for movies
and television programs? And why would it
be possible to sell video programs for infants
with names like Baby Einstein?2 How might
1 Broadhurst, Cunnane, & Crawford, 1998.
2 © The Disney Corporation.
we manipulate the environment to improve
intelligence?
There are two ways this question can be
interpreted. The less interesting interpreta¬
tion is “Can the environment be manipu¬
lated to improve test scores?” A more inter¬
esting question is “Can the environment
be manipulated to improve general mental
competence?” The answer to both questions
is “yes.”
Environment is a catch-all term. In dis¬
cussing environmental effects on intelli¬
gence it is useful to make a distinction
between the physical environment and the
social environment. The physical environ¬
ment involves things like nutrition, air pol¬
lutants, and disease - anything that makes
itself felt by direct physical action. The social
environment involves things like education,
social actions that enhance or threaten secu¬
rity, and opportunities for self-development
of cognitive skills. Both the social and the
physical environment alter the brain’s activ¬
ity and, as Shakespeare said, infect (leave
a physical trace on) the brain. It does not
always pay to go to this level of analysis.
A teacher in an elementary school does not
257
258
HUMAN INTELLIGENCE
care whether a vocabulary-learning exer¬
cise will move brain activity related to
word recognition from the frontal cortex to
the posterior temporal cortex. The teacher
wants to know if the exercise will improve
the way the students use words. It is impor¬
tant to keep our discussions at the appropri¬
ate level of analysis - Brunswikian symmetry
again!
9.1. Three Key Issues in the Study of
Environmental Effects
In much of the literature environmental
effects are contrasted to genetic effects. This
is best exemplified by the continuing argu¬
ments over genetic versus environmental
causes of intelligence. A more sophisticated
way of looking at this debate is to determine
the extent to which genetic and environ¬
mental constraints limit behavioral poten¬
tial. This is not a simple task. Three issues
have to be kept in mind. They are the con¬
cepts of reaction range; the distal-proximal
distinction, and the problem of collinear-
ity. While these issues have been discussed
before, they assume special importance in
the study of environmental effects.
9.1.1. Reaction Range
The concept of reaction range was intro¬
duced in Chapter 1, section 1.5. To review
briefly, genetics determines a potential for
the expression of a trait, but in most situa¬
tions the extent to which the trait is actu¬
ally expressed [if at all) is influenced by the
environment. At one extreme we have eye
color, for which there is virtually no reac¬
tion range; at the other extreme are traits
like alcoholism and Alzheimer's dementia,
where a person inherits a risk that can be
triggered by environmental events. Intelli¬
gence is decidedly the latter type of trait.
We can think of the environment as
determining where a given person will oper¬
ate, within his or her genetically specified
reaction range. Because different people
will operate within different environments,
variations in behavior will be determined
both by differences between people in reac¬
tion range and differences in environmental
influences. This is the reason that h2 varies
in different situations; it reflects the relative
importance of differences in reaction ranges
and differences in environmental conditions
that operate within the potential afforded
by reaction ranges.
Environmental effects are often illus¬
trated by experiments in which two rela¬
tively extreme environments are contrasted
with each other. Such an experiment shows
what might happen under certain extreme
conditions. While this information can be
important, it does not tell us what is likely
to happen under normal conditions of envi¬
ronmental variation. At this point we have
to deal with a measurement issue. In order
to define the extent of environmental varia¬
tion we have to have a metric specifing how
close two environments are to each other.
At present no such metric exists, largely
because we do not have a good theory of
environmental variation. In fact, we have
hardly any such theory at all.
9.1.2. Proximal and Distal Causation
As any parent of two or more children
knows, people exert a lot of influence over
their own environments. Where one person
has experiences, another will gain knowl¬
edge. Recall Ackerman's emphasis on the
importance of intellectual engagement upon
the development of intelligence in adult¬
hood (Chapter 4). On the other hand,
environments vary in the extent to which
they encourage the acquisition of skills and
knowledge. For instance, it is well estab¬
lished that parents in the higher socioeco¬
nomic status (SES) range are more likely
to encourage their children to engage in
exploratory and problem-solving activities
than are parents in the lower SES ranges.3
Suppose that variations in the initial ten¬
dency to explore an environment have a
partially genetic basis. How are we to inter¬
pret a study that demonstrates that children
show greater cognitive development if they
5 Nisbett, 2009.
ENVIRONMENTAL EFFECTS ON INTELLIGENCE
259
Low-quality
cognition
Figure 9.1. The design of a hypothetical experiment demonstrating
how the development of cognition may depend on environments
that encourage exploration. The environment acts as a proximal
cause of the level of cognition. Unknown to the experimenter, the
participating children are genetically divided into high and low
explorers. Exploration tendency acts as a distal variable, producing
variation in cognitive development within one environment but not
within the other.
are exposed to enriched environments that
encourage exploration? Would this be a dis¬
tal genetic or proximal behavioral effect?
In fact, many such studies have been
conducted. Figure 9.1 shows a widely
used design. Children (of unknown genetic
potential for exploration) are randomly
assigned to an environment that either
encourages or restricts cognitive develop¬
ment. We suppose that, unknown to the
experimenter, the children vary in their
genetic tendency to explore. To take an
extreme, assume that there are geneti¬
cally determined “high” and “low” explor¬
ers. Following the convention introduced in
Chapter 1, the distinction is shown in ellip¬
soids, to designate an unobserved quantity.
The experimenter can observe the sort of
environment that the children are exposed
to (center rectangles) and their cognitive
performance (right-hand rectangles). If cog¬
nitive performance is better in the encourag¬
ing than in the restricting environment, the
experimenter is justified in concluding that
the environments influenced cognition, act¬
ing as proximal variables. However, genetic
potential has also had an influence, as a
distal variable. The restrictive environment
does not permit genetic influences to act;
the encouraging environment does; and the
high explorers take more advantage of this
opportunity than do the low explorers.
Figure 9.1 presents the proximal-distal
distinction as it might occur in an exper¬
iment, where the distinction is clear-cut.
The situation can be much more confusing
260
HUMAN INTELLIGENCE
in the natural world. The extent to which
children are raised in an environment that
encourages exploration varies with socio¬
economic status (SES). Generally, low SES
children are raised in more restrictive envi¬
ronments than are the children of upper
and middle SES families. Upper and mid¬
dle SES children also, on the average, have
higher IQ scores than lower SES children.
But it may be that the children in upper- and
middle-class SES families are more likely to
be genetically predisposed to explore than
are lower SES children. It might be that
the upper- and middle-class SES parents are
genetically predisposed to interact with chil¬
dren in a way that encourages exploration.
And there are other possibilities. It could be
that higher SES families are subject to fewer
social stresses, and therefore are better able
to create an encouraging environment. Out¬
side of the laboratory proximal and distal
variables are often thoroughly mixed up. In
interpreting studies of the causes of intelli¬
gence this caution should be kept in mind.
9.1.3. Collinearity
Collinearity refers to a situation in which
several possible causes of a phenomenon are
themselves correlated. Collinearity presents
serious difficulties for anyone interested in
determining environmental influences on
cognition. To illustrate, there is a nega¬
tive correlation between family income and
children's test scores. Nisbett has suggested
a simple economic solution to the prob¬
lem: providing poor people with subsidies
to improve their and their children's life
situation.4 However, he has also pointed out
that this would not help immediately, for
family income is correlated with a number
of other potential restrictions on children’s
development, including nutrition and child-
rearing practices. Nisbett does not stress the
point, but SES, which includes education
and income, is also correlated with genetic
inheritance, for people tend to marry people
of their own general social class, especially
4 Nisbett, 2009, p. 192.
with respect to education.5 Providing poor
people with a subsidized income for child
rearing might improve nutrition and certain
other aspects of the environment immedi¬
ately, might improve child-rearing practices
over generations (a point Nisbett stresses),
and would probably have little effect on
genetic makeup.
The same thing happens on an interna¬
tional basis. Across nations, poor nutrition is
associated with low test scores. Poor nutri¬
tion is most likely to occur in countries that
have poor school systems, and to attack chil¬
dren whose parents have low IQ scores. So
what is causing what?
Look at the issue symbolically. Let I be
intelligence and Q . . . C* be a set of pos¬
sible causes of intelligence. We observe a
correlation between I and Q. But Q is
correlated with many of the other possible
causes, C2 . . . C^, so the observed I, Cx cor¬
relation might be due to any of the other
possible causes. And to make things just a
little harder, we have to consider possible
feedback mechanisms. By any conceptual
definition, intelligence is adaptive behavior.
People with high intelligence are, in gen¬
eral, able to cope with unfavorable phys¬
ical and social environments more effec¬
tively than people of low intelligence. To
take a not unrealistic example, driving while
intoxicated puts you at risk of severe head
injury, which cannot be a good thing for
your intelligence. But if you are intelligent,
you are less likely to drive while intoxicated,
compared to an unintelligent person.
The problem of collinearity is not unique
to studies of intelligence. However, it is
unusually severe in this field. For ethical
and practical reasons, it is seldom possible
to avoid the collinearity problem by con¬
ducting controlled experiments, where the
levels of various causes are manipulated by
an experimenter. It situations where con¬
trolled experiments can be conducted, there
is often reason to question whether the lab¬
oratory results can be extrapolated to the
day-to-day situation. The same criticism can
5 Blackwell & Lichter, 2000.
ENVIRONMENTAL EFFECTS ON INTELLIGENCE
261
be levied against simple regression mod¬
els, where one variable is “held constant”
by statistical means. In these studies inves¬
tigators determine the influence of causal
variable Q on the residual variance in intel¬
ligence (I) after removing variation in I due
to all other causal variables of interest. This
technique is a useful way of determining
how much of the variation in the intelli¬
gence test scores can be associated with each
of the predictors. We have to remember that
the statistical model implies that each causal
variable can be manipulated independently.
In practice this may not be possible. Modern
studies rely heavily on a structural equation¬
modeling technique similar to those used in
quantitative behavior genetics, as described
in the last chapter. This can go a long
way toward disambiguating collinearities.
Unfortunately, the resulting models become
complex.
Given all these analytical problems,
understanding environmental effects on
intelligence is going to be hard work. Will
the results be worth the effort? Only if envi¬
ronmental effects are very large. The next
section shows that they can be.
9.2. How Much Can Environments
Matter? The Cohort (Flynn) Effect
In Chapter 8 we saw that the existence
of substantial heritability coefficients, h2,
showed that there are genetic influences
on intelligence, without revealing anything
about how the genetic influences operate.
There is an analogous situation with regard
to the environment. People are growing
smarter at a rate that is far greater than can
be accounted for by genetic/e volution ary
effects. Senior citizens may find this hard to
believe (unless they are talking about their
grandchildren], but intelligence test scores
rose throughout the twentieth century. This
is called a cohort difference, where a cohort
refers to the group of individuals born at a
particular time - for example, the cohort of
people born in 1970. Throughout most of the
twentieth century each successive cohort
was more intelligent than its predecessors.
The beginnings of studies of the cohort
effect are interesting, because it appears
the researchers expected to find something
quite different from what they found.
9.2.1. The Cohort Effect
In the 1940s Read Tuddenham, a professor
at the University of California, compared
intelligence test scores for White men who
had enlisted in the US Army during World
Wars I and II.6 Tuddenham wanted to test
R. B. Cattell's conjecture that intelligence
should drop over time, because people with
high intelligence test scores have fewer chil¬
dren than people with low test scores.7
Tuddenham found just the opposite. During
the twenty-five year period between World
War I (American participation 1917-18] and
World War II (American participation 1941—
45], the mean intelligence test score of young
White males in the United States increased
by approximately one standard deviation
unit. See panel 9.1 for more details of Tud¬
denham 's work.
In the 1950s K. Warner Schaie, who had
worked with Tuddenham as an undergrad¬
uate, began graduate studies at the Uni¬
versity of Washington. His thesis advisor
was Charles R. Strother, a well-known clin¬
ical psychologist interested in aging. At the
time a puzzling anomaly had been noticed.
Studies of aging used one of two designs.
In a cross-sectional design people of differ¬
ent ages are tested at the same time. Thus
a study in, say, 1955 might involve people
who were in their twenties, thirties, for¬
ties, and so on. In a longitudinal study a
group of people of more or less the same
age is followed for some period of time. For
instance, an investigator might begin a study
in 1955, with participants in their twenties,
and follow them as they aged over the next
twenty to thirty years. Cross-sectional stud¬
ies led to the conclusion that cognitive func¬
tions begin to decline after a peak in the
late twenties, with some differences in tim¬
ing for different functions. Studies using
6 Tuddenham, 1948.
7 Cattell, 1940.
262
HUMAN INTELLIGENCE
Panel 9.1. The Tuddenham Study
In 1917 the United States Army devel¬
oped the Army Alpha Test as a device
for screening recruits. In World War
II (American participation 1941-45] the
Army used a successor test, the Army
General Classification Test. In order to
compare the two tests the Army gave a
version of the World War I test to 768
World War II soldiers, selected to repre¬
sent the demographics of World War II
enlistees. The median score of the World
War II soldiers was approximately one
standard deviation higher (using World
War I norms) than the median score for
the World War I soldiers.
Tuddenham pointed to several possi¬
ble causes of the discrepancy. He felt that
the most important of these was that the
1945 soldiers had, on the average, much
more education than the 1917-18 soldiers.
He reinforced this conclusion by showing
that those World War I soldiers who were
literate had test scores reasonably close
to the scores of the World War II sol¬
diers. Figure 9.2 shows both the general
effect and the effect of introducing liter¬
acy as a covariate. In many ways Tudden-
ham’s study anticipated discussions ol the
cohort effect that were to emerge over
the next fifty years.
longitudinal designs found that declines did
not begin until people were in their fifties or
sixties.
Schaie and Strother realized that both
cross-sectional and longitudinal studies are
confounded with cohort effects, but in dif¬
ferent ways.8 A cross-sectional study com¬
pletely confounds age with cohort; if people
are tested in, say, 2000, then all the thirty-
year-olds will have been born in 1970, the
forty-year-olds in i960, and so forth. A lon¬
gitudinal study is conducted within a single
cohort; if people are followed from twenty
to fifty, beginning in 1980, all participants
will have been born in i960. Schaie and
Strother offered an elegant solution to the
design issue, the cohort-sequential design.
In a cohort-sequential design a cross-
sectional sample of people of varying ages
(and birth cohorts) is collected at time 1.
Then, after a period of time (in Schaie
and Strother's case, seven years) a second
cross-sectional design is conducted, drawing
a sample from the same population, at time
2. In addition, the first sample is contacted
again, and as many people as possible are
retested in a second wave. The procedure
is repeated over as many testing waves as
8 Schaie & Strother, 1968.
possible. Cohort and age studies can now be
analyzed separately.
Suppose that two test waves have been
conducted, in 1956 and 1963 (the actual
test dates of Schaie and Strother’s first two
waves). Consider the following four partici¬
pants:
Person A bom 1936 Tested in 1956 at age
twenty and in 1963 at
age twenty-seven
Person B bom 1916 Tested in 1956 at age
forty and in 1963
at age forty-seven
Person C bom 1943 Tested .
in 1963 at age twenty
Person D bom 1923 Tested .
in 1963 at age forty
There are two longitudinal contrasts:
between person A in 1956 and 1963, covering
the age span twenty to twenty-seven, and
between person B in 1956 and 1963, cover¬
ing the age span forty to forty-seven. There
are three cross sectional contrasts. They are
between person A and B, tested in 1956, at
ages twenty and forty; between persons A
and B, tested in 1965, at ages twenty-seven
and forty-seven, and between persons C and
D, tested in 1963, at ages twenty and forty.
There are two cohort contrasts: between per¬
son A, tested in 1956, and person C, tested in
ENVIRONMENTAL EFFECTS ON INTELLIGENCE
263
■ All WW I
■ Literate WW I
■ WW 2
Figure 9.2. Scores on the Army Alpha Test obtained by World War
I soldiers, World War II soldiers, and literate World War I soldiers.
Scores are shown by increasing deciles - that is, the scores on the
far left are for the tenth decile of each group, and so forth. Data
from Tuddenham, 1948.
1963, both at age twenty at the time of test¬
ing, and between person B, tested in 1956,
and person D, tested in 1963, both at age
forty at time of testing.
The cross-sequential design makes it
possible to determine longitudinal, cross-
sectional, and cohort effects. It does con¬
found date of testing with other effects, but
I have never heard of any reason that date of
testing, alone, should have an influence on
test scores.
In addition to its design aspects, Schaie
and Strother’s study was noteworthy for
its sophisticated measurements. Instead of
relying on a single test score, Schaie and
Strother used numerous tests to evaluate
different aspects of intelligence. For our pur¬
poses the most interesting are their measures
of inductive reasoning, verbal comprehen¬
sion, and spatial reasoning. These measures
correspond well to the modern g-VPR the¬
ory: inductive reasoning maps onto g; ver¬
bal comprehension maps onto the verbal
dimension; and the spatial reasoning task
they used appears to me to be a mix of
the perceptual and rotational dimensions in
Johnson and Bouchard’s VPR model.9
9 Johnson & Bouchard, 2005.
The 1968 article reported large cohort
differences in all three types of reason¬
ing. They were invariably in favor of later-
born cohorts, which is consistent with
Tuddenham’s results. Schaie and Strother
concluded that the rapid drop in intelligence
after early middle age (thirty to forty) that
had been observed in cross-sectional stud¬
ies was very largely a cohort rather than an
aging effect.
Schaie has continued and expanded the
study, using the cross-sequential design, col¬
lecting data every seven years for over forty
years! The study, which is now known as
the Seattle Longitudinal Study, is described
in panel 9.2.
Figure 9.3 shows the effect of cohort
changes since the birth cohort of 1903 (who
were fifty-three at the time of the first
testing) until the last testing in 1998. The
results are typical of what is found in this
field. Compared to the birth cohort of
1903, all subsequent birth cohorts showed
a steady increase in reasoning. The increase
in spatial orientation ability was somewhat
more uneven, but still marked. The abil¬
ity to extract verbal meaning increased until
the birth cohort of 1952, and subsequently
decreased.
264
HUMAN INTELLIGENCE
Panel 9.2. The Seattle Longitudinal
Study
The Seatde Longitudinal Study grew out
of a fortuitous combination of interests,
combined with K. Warner Schaie’s tena¬
cious pursuit of a good idea. Follow¬
ing his undergraduate work with Read
Tuddenham at the University of Califor¬
nia, Berkeley, Schaie moved to the Uni¬
versity of Washington to begin studies
for his Ph.D. with Charles R. Strother,
a senior professor in clinical psychology
who was very interested in public health
issues.
Strother had helped found the Seattle
Group Health Program, one of the first
large, open-access health maintenance
organizations fHMOs) in the United
States. (Previous HMOs had been asso¬
ciated with certain industries or institu¬
tions.) He and Schaie realized that the
people enrolled in the program consti¬
tuted a group that could be approached
over time. With the cooperation of the
Group Health organization, Schaie has
managed to collect data over seven con¬
secutive waves, extending until the latest
wave in 2005. This has been an impressive
feat of scientific management and persis¬
tence.
The project has measured intelligence
using tests based on Thurstone’s Primary
Mental Abilities model, which includes
separate factors for verbal reasoning,
nonverbal reasoning, induction, and
numerical/arithmetic reasoning. Data has
also been gathered on personality fac¬
tors, health, and lifestyle. Related stud¬
ies have been conducted on interventions
designed to ameliorate cognitive deterio¬
ration in old age. The results have been
reported in two books* and numerous
research papers.
* Schaie, 1996, 2005.
To make these numbers a bit less abstract,
here is a hypothetical example. Since the
measures of reasoning are closest to what
would be considered measures of g in mod¬
ern theories, the comparisons will be made
for the reasoning test, converted from stan¬
dard deviation units to IQ points.
Consider two people of age thirty, born
in 1903 and tested in 1933. We select them so
that one is at the median level of intelligence
in his or her 1903 birth cohort and the other is
at the eighty-fifth percentile. By definition,
the first person will have an IQ of 100 and
the second an IQ of 115, by 1933 standards.
Birth year of cohort, starting in 1903 (03)
— •#— Reasoning
— ■ - Spatial orientation
•••if Verbal meaning
Figure 9.3. The increase in three dimensions of intelligence over
cohorts, using the 1903 birth cohort as a base. Data from Schaie,
2005, p. 135; converted to standard scores.
ENVIRONMENTAL EFFECTS ON INTELLIGENCE
265
We now take two more people, age thirty,
but born in 1973, who are tested in 2003, and
who are at the fiftieth and eighty-fifth per¬
centile by 2003 standards. They would have,
respectively, IQs of 115 and 130 by 1933 stan¬
dards! People are getting smarter.
9.2.2. The Cohort Effect Goes
International l The Flynn Effect
Early in the 1980s Jim Flynn, a political
philosopher at the University of Otago, in
New Zealand, became interested in changes
in the difficulty of standardized tests of intel¬
ligence, such as the Wechsler tests and the
Stanford-Binet tests. Flynn located seventy-
three different studies, covering eighteen
pairings in which the same people had been
given both a current test and a test stan¬
dardized at some earlier time (e.g., the orig¬
inal Wechsler test, standardized in 1935—
37, and the second version, standardized in
1953- 54). h1 seventeen of the eighteen com¬
parisons people scored higher on the ear¬
lier version of the test than on the later
version.10 For instance, in the comparisons of
the original and second Wechsler tests there
was a difference of 4.69 IQ points, approx¬
imately one-third of a standard deviation
unit, over a period of seventeen years. Flynn
concluded that whatever trait the Wechsler
tests measure had increased steadily in the
US population.11
Flynn then examined test scores in stan¬
dardization studies conducted in fourteen
different industrialized nations.12 All of
them showed cohort effects, ranging from a
high of twenty-five IQ points (France, 1949-
74) to a low of six points (United States,
1954- 78).
In subsequent studies Flynn found that
the increase in scores over time was
10 The one negative finding involved comparisons of
tests intended for populations of different ages.
11 Flynn, 1984.
12 Flynn, 1987. The nations involved were Austria,
Australia, Belgium, Canada, France, German
Democratic Republic (now part of Germany),
German Federal Republic (now part of Ger¬
many), Great Britain, Japan, the Netherlands, New
Zealand, Norway, Switzerland, and the United
States of America.
markedly larger for nonverbal, culture-
reduced tests than it was for verbal tests.
This is consistent with results from the
Seattle longitudinal study, where the cohort
effects on a nonverbal reasoning test were
greater than those on a verbal compre¬
hension test (Figure 9.3). Flynn also con¬
trasted the increase in US intelligence test
scores with decreases in SAT scores over
the period since World War II. 13 He con¬
cluded that, at a descriptive level, there had
been substantial gains in IQ scores, intel¬
ligence in the narrow sense, over the first
three quarters of the twentieth century, and
that they were concentrated in abstract rea¬
soning tests. He also expressed considerable
skepticism about whether intelligence in the
conceptual sense had similarly increased, for
he did not believe that people today are
markedly smarter than their ancestors.
9.2.3. Further Documentation
of Flynn's Observations
Flynn’s research caught the public eye. Her-
rnstein and Murray coined the term “Flynn
effect” in 1994. 14 In 1996 the American Psy¬
chological Association sponsored a sympo¬
sium on possible causes of the effect.15 Flynn
himself published a book in which he con¬
sidered further studies and their implica¬
tions in 2007. 16 Many articles by other writers
have commented on the effect. Surprisingly,
only minimal references have been made to
Schaie's work, even though it was very well
known to gerontologists.
Research on the Flynn version of the
cohort effect has been greatly helped by a
social phenomenon. Although only a few
13 The drop in scores was dramatic, especially over the
period 1970-92. However, the contrast over cohorts
is problematical. The SAT is taken by a self-selected
population, people intending to enter college. Dur¬
ing the 1950-2000 period there was a considerable
increase in the percentage of high school students
who took the test. Therefore, it is hard to draw con¬
clusions about changes in the cognitive competence
of the US population, in general, from changes in
SAT scores over time.
14 Hermstein & Murray, 1994.
15 Neisser,i998.
16 Flynn, 2007.
266
HUMAN INTELLIGENCE
developed countries currently have compul¬
sory military service, several European coun¬
tries require that young men register for
possible conscription in case of national
emergency. When they enroll the registrants
are given medical and psychological exami¬
nations, including a test of cognitive skills.17
Raven’s Standard Progressive Matrices or
similar tests are used by some countries. As
a result, it is possible to compare scores of
virtually the entire population of eighteen-
year-old men, on the same test, over the
years.
Figure 9.4 shows the test scores for Dan¬
ish registrants, as a function of time of
testing. 18 This figure, which is typical of
data obtained in other studies, shows three
important features of the Flynn effect. The
first is that the effect is certainly there;
test scores increased markedly in the thirty-
year period from 1958 until 1988. The sec¬
ond is that the effect is greatest at the low¬
est level of scores. The two top levels, the
ninetieth and seventy-fifth percentiles, show
very little effect.19 The third is that there is
some indication that the effect is slowing
over time. Studies using data obtained since
1990 show considerably smaller increases,
and some have even suggested decreases in
the 1980s and 1990s birth cohorts.20
9.2.4. What Does the Cohort Effect
Tell Us?
Does the cohort effect exist? Objections can
be raised to the design of many of the studies
of the effect, individually. Some of the prob¬
lems are discussed in panel 9.3. These prob¬
lems raise potential objections, rather than
disproving the findings. Also, the nature of
17 The United States currently requires registration
but does not do any testing.
18 Teasdale & Owen, 2000.
19 It has been claimed that the lack of effect at the
top is due to an artifact in scoring called the “ceiling
effect.” If a substantial number of people score at the
top, their scores cannot increase (J. Raven, 2000).
This argument applies only to the highest percentile
considered, here the ninetieth percentile. As the
figure shows, there is clearly a trend to smaller and
smaller cohort effects as the percentile increases.
20 See Teasdale & Owen, 2008, for data and further
references.
the objections varies from study to study.
This does not seem to matter; the results are
surprisingly robust. Test scores rose substan¬
tially from the 1920s until the late 1980s, and
then either leveled off or, possibly, dropped.
The facts are clear. What to make of them
is not.
Are people really getting smarter? If the
correlations between intelligence test scores
and measures of socially important cogni¬
tive achievements, such as school grades
and income, were close to one, the answer
would have to be “yes." If the correlations
were close to zero, changes in test scores
would be irrelevant to discussions of cog¬
nition in everyday life. In fact, the correla¬
tions are in the .4~.6 range; certainly not one,
but high enough to show that test scores
have to be taken seriously. (More details
are given in Chapter 10.) As was pointed
out in Chapter 1, what correlations in this
range show is that intelligence tests evalu¬
ate “life-relevant” cognitive skills and some
skills that are unique to the testing situa¬
tion. Is the cohort effect due to changes
in the life-relevant skills or the test-unique
skills?
The test sophistication argument asserts
that the cohort effect is due to an increase
in test-unique skills, because over the years
cognitive testing has become more a part of
everyday experience. I do not think that this
argument can be maintained. If test scores
increasingly reflected test-taking skills, then
the correlation between test scores and other
measures of cognitive achievement should
have fallen over time. There is no evi¬
dence that this is the case. Also, American
test scores continued to rise in the 1960-80
period, which is well after the widespread
introduction of testing in the United States.
The rise in test scores almost certainly rep¬
resents a rise in cognitive skills across the
developed world.
The cohort effect varies with the type
of test. By far the greatest rise is found
in nonverbal, g-loaded tests, such as pro¬
gressive matrix tests. Flynn21 contrasts this
rise to the stability in performance on
21
Flynn, 1987; 2007, Chapter 2.
ENVIRONMENTAL EFFECTS ON INTELLIGENCE
267
Secular Trends
Test Date 1958 1968 1978 1988 1998
Figure 9.4. Progressive matrix test scores for Danish
eighteen-year-old men registering for military enlistment,
1958-98. From Teasdale 81 Owen, 2000, with permission from
Elsevier.
assessments of explicitly taught school sub¬
jects. IQ test scores rose from 1971 to 2002,
especially on tests requiring abstract reason¬
ing. The twelfth-grade achievement scores
for the National Assessments of Educa¬
tional Progress (NAEP) tests, which focus
on school subjects, did not rise. Flynn con¬
cluded that there has been an increase in
abstract reasoning skills without a concomi¬
tant increase in academic knowledge. In
terms of the three-stratum model, Gf is up
while Gc is flat. Why?
As Flynn and others have pointed out, any
discussion of the cohort effect has to deal
with a paradox. Flynn estimated that there
has been a rise of .3 IQ point per year. If
we project this backward, to calculate the
intelligence of past cohorts, we find that
the mean IQ score for young male adults
in 1942 was equivalent to 80 by 2008 stan¬
dards. If we accept this finding at face value,
it leads to the conclusion that almost half
of the soldiers who fought in World War II,
the group that have been called “The Great¬
est Generation,”22 would not meet the men¬
tal requirements for enlistment into today’s
army. This conclusion is ridiculous. But
what has happened?
To answer that question we have to ask
what might cause the cohort effect in the
first place. The cohort effect cannot be due
to genotypical changes. There is a negative
22 Brokaw, 1998.
z68
HUMAN INTELLIGENCE
Panel 9.3. Design Issues in the
Study of Cohort Effects
Studies of the cohort effect have taken
two forms. The first, which is typified
by the Tuddenham study, some aspects
of the Seattle Longitudinal Study, the
normalization comparisons for Raven’s
Standard Progressive Matrices, and the
various European studies of enlistees,
compares the scores obtained on the same
test by people of the same age, but from
different cohorts - for example, regis¬
trants for the military in Denmark in 1958,
and then again in 1978. A generalization is
then drawn to some larger population. 1
will call this the same-test , different-cohorts
paradigm. Drawing conclusions from this
design requires two assumptions.
The first is that the test is a meaning¬
ful way to evaluate cognition in different
cohorts. This seems a reasonable assump¬
tion for “culture-reduced” tests, such as
progressive matrix tests, given to the
same overall population - for instance,
the population of the United States in
1933 compared to the population in 1983.
It is also reasonable if the cohorts are
not far apart in years, for cultures do not
change that quickly.
There are situations where this as¬
sumption would not be reasonable. For
instance, if the tests were to be con¬
ducted in a developing country it might
be the case that proportionately more
people in the more recent cohort would
be accustomed to the testing paradigm,
due to dramatic increases in schooling,
and with it, testing. Such an argument
is much less tenable for a comparison
of cohorts in an industrially developed
country.
Tests of verbal behavior and general
knowledge have to be modified for the
cohort involved. To take an extreme,
albeit slightly frivolous, example, in 3938
the term gay bachelor referred to an
unmarried man who enjoys the company
of women. By 2008 the term had come
to acquire a rather different meaning.
Any test involving cultural knowledge
also faces the danger of being frozen in
time or restricted to a particular cultural
group. In order to maintain widespread
applicability, commercial tests that eval¬
uate crystallized knowledge (Gc) tend
to evaluate “least common denomina¬
tors,” knowledge that is held widely
through the society. Common knowledge
changes over cohorts. The tests have to be
changed accordingly.
The second assumption is that the
cohorts are similar samples of a larger
population. Findings on the cohort effect
have been unhesitatingly generalized to
entire populations. In fact, few studies
utilize a random sample of the popula¬
tion for which generalization is intended.
For instance, the 1992 United States stan¬
dardization sample for the widely used
Raven’s Standard Progressive Matrices
test was entirely drawn from Des Moines,
Iowa. To what extent, then, can we gen¬
eralize to the population of the United
States?* Tuddenham compared US mili¬
tary recruits from World War I to recruits
from World War II. Recruitment proce¬
dures were not the same in the two wars,
so is it valid to make an inference about
changes in population IQ? The Seattle
Longitudinal Study drew from a popula¬
tion of enrollees in a health care program.
To what extent is this population repre¬
sentative of the US population in general?
Also, to what extent did the nature of
the enrollees in the health care program
change over the almost half-century life¬
time of the project?
Because of questions like these the
European military registration studies
are particularly valuable. They involve
repeated sampling of the same subpop¬
ulation, young men eligible for military
service, over fairly brief time intervals.
The fact that a cohort effect appears in
several European studies of this group is
an important confirmation of other same-
test, different-cohort designs.
ENVIRONMENTAL EFFECTS ON INTELLIGENCE
269
Flynn relied on a second paradigm, in
which two tests, with norms established
at different times, were given to the same
sample. Call this the two-test, one-cohort
design, and refer to the test with the ear¬
lier norms as Test 1 and the second test as
Test 2. Suppose that scores are higher, in
terms of percentiles defined by the origi¬
nal standardization, on Test 1 than on Test
2. This implies that the standardization
sample for Test 1 had lower abilities than
the standardization sample for Test 2.
The argument depends on the two
standardization samples being equally
representative of the general population
at the time that the standardization is
done. Considerable care is taken in stan¬
dardizing tests such as the Stanford-Binet
and Wechsler tests, as these tests are
widely used in clinical practice, and to
establish legal competency and/or qual¬
ification for special education programs.
The cohort effect can be illustrated with
these tests.
* Raven, 2000. Des Moines was chosen because,
on some statistical criteria, such as age distri¬
bution and distribution of ethnic groups, the
city was claimed to have matched the United
States as a whole. The same logic was used
for the British standardization, where the sam¬
ple was drawn from Dumfries, Scotland. The
problem with this approach is that the match¬
ing is solely on those variables that the inves¬
tigator thinks are appropriate, and leaves any
other measures free to vary. For instance, in
the United States educational standards vary
widely across states and even across school dis¬
tricts within states. Was the quality of educa¬
tion in Des Moines equivalent to the typical
quality of education in the United States at the
time? Random sampling avoids such problems
by equating statistical expectations for all covari¬
ates, not just for those felt to be important by the
investigator.
correlation between IQ scores and fertil¬
ity [number of live births per woman).23
Because of this, some investigators have
argued that on a population basis the genetic
potential for intelligence has dropped, even
though IQ scores have risen.24
While this argument has a certain amount
of merit, it is not completely compelling.
Differential fertility, alone, does not guar¬
antee a "dysgenic effect,” lowering the mean
intelligence of the population by lower¬
ing the genetic potential. This result would
depend upon a number of other things,
including the initial relative frequencies of
various genetic potentials and the mating
practices in the society.25 One could create a
situation in which the genetic potential of a
population increased, even though there was
a negative correlation between intelligence
and fertility. However, this can happen only
if the society follows mating practices that
are not found in any of the industrialized
nations where the cohort effect has been
23 Herrnstein & Murray, 1994.
24 Herrnstein & Murray, 1994, pp. 348-352; Lynn, 1998.
25 Loehlin, 1998; Preston, 1998.
observed.26 The cohort effect is environ¬
mental.
There is an analogy between the cohort
effect and the finding that the heritability
coefficient is .50 or higher (Chapter 8). The
heritability finding shows that genetics is
important, but it does not identify the rele¬
vant genes. The cohort effect does the same
thing for environmental influences. It shows
that environmental influences are impor¬
tant, but does not identify the influences.
If we look at any time-linked phenomenon
extending over a period of seventy to one
hundred years, we find that there have been
26 This situation could arise if males with high genetic
potential had more offspring than males with lower
genetic potential. The discrepancy would have to be
large enough to overcome the negative correlation
between intelligence and female fertility. This could
happen in a polygamous society, if the males with
high genetic potential had more access to women
than males with low genetic potential. There may
be one spectacular case. Genetic analyses indicate
that 8% of the men living in Central Asia (approxi¬
mately .5% of the world's men) are descended from
one individual who lived approximately a thousand
years ago. If we combine this observation with his¬
torical records, suspicion falls on Genghis Khan, the
founder of the Mongol Empire (Zerjal et al., 2003).
270
HUMAN INTELLIGENCE
so many, highly collinear social changes that
it is virtually impossible to single out any
one of them and say “That’s it.”
The analogy goes further. Behavioral
geneticists and biomedical scientists have
found some genes that are very bad for intel¬
ligence, but have not been able to isolate the
genes that produce high intelligence. We can
identify environments that are very bad for
intelligence, but have not been able to iden¬
tify environments that are very good for it.
9.3. The Physical Environment
Several features of the physical environment
are associated with low intelligence test
scores. These include poor health practices,
frequent or severe injuries, substance abuse,
poor nutrition (especially in infancy), and
exposure to atmospheric pollutants. In some
cases the direction of causality is clear-cut.
Any physical agent that damages the struc¬
ture of the brain or interferes with brain pro¬
cessing may reduce intelligence, on either
a temporary or a permanent basis. That is
simply common sense. In other cases there
is a statistically reliable association between
intelligence and some potentially harmful
agent, but causality is difficult to determine.
One of the reasons for this is a sort
of chicken-and-egg problem: which came
first, reduced intelligence or exposure to the
agent? Linda Gottfredson, who has written
widely on the social implications of intelli¬
gence, has pointed out that the task of fol¬
lowing health and safety regimens is itself
cognitively challenging.27 To take a specific
case, intelligence test scores are predictors
of involvement in motor vehicle accidents,
and such accidents are one of the common¬
est causes of head injury.25 Granted that
individuals who have suffered head injuries
behave unintelligently at times, how likely
were they to behave unintelligently prior to
the accident?
Collinearity makes it hard to determine
causality. Exposure to dangerous conditions
27 Gottfredson, 2007b.
28 O’Toole, 1990; Smith & Kirkham, 1982.
and low intelligence test scores are corre¬
lated with many other variables, such as
socioeconomic status (SES), low parental
intelligence, and inadequate access to good
schools, that could themselves adversely
affect intelligence. Picking out just one of
these factors as the cause of low intelligence
becomes hard to justify.
In order to assert that a particular incident
or environmental condition has affected
intelligence it is necessary to show (a) that
low intelligence did not contribute to the
incident or condition, and (b) that other
possibly relevant conditions have been mea¬
sured or controlled experimentally. In the
ideal case pre-morbid measures of intelli¬
gence, taken before the incident or expo¬
sure, are compared to measures taken after
the fact.
9.3.1. Direct Insults to the Brain
Chapter 7 discussed the relationship bet¬
ween brain structures and intelligence.
There it was pointed out that damage to
certain areas of the brain will result in loss
of intelligence and, if the injury is suffi¬
ciently discrete, may disrupt certain intel¬
lectual functions and not others. Damage to
the forebrain-parietal system will disrupt the
working memory-attentional control com¬
plex, thus damaging the functions underly¬
ing general reasoning (Gf or g ). Damage in
the medial temporal region, and especially
to the hippocampus, will disrupt the ability
to form new memories ( anterograde amne¬
sia). The disability can be severe enough
that the affected person has to be in custo¬
dial care for life. Paradoxically, some writers
have claimed that this does not affect intel¬
ligence, on the ground that IQ test scores
many not be lowered. Because any reason¬
able definition of intelligence has to include
the ability to learn from experience, the fact
that profound anterograde amnesia does not
influence IQ test scores simply shows that
test scores are only partial indices of intelli¬
gence.
Less dramatic, but important, losses of
cognitive functions can occur when the
brain is subjected to apparently minor
ENVIRONMENTAL EFFECTS ON INTELLIGENCE
271
physical damage. Closed head injury, con¬
cussion , is of special interest because of its
prevalence; there are more than a million
cases annually in the United States.29 Closed
head injuries are often followed by a period
of disorientation that is obvious both to the
individual and to others. In most cases this
subsides. When conventional intelligence
tests are given, a year or more later, effects
are often not found. When effects are found
they are usually on abstract, nonverbal rea¬
soning. Tests that emphasize verbal cogni¬
tion (or Gc, if you wish to use the Gf-Gc
model) appear to be much less sensitive to
the aftermath of severe concussions. So, if
we were to restrict ourselves to the intel¬
ligence test data alone, we might conclude
that “a knock on the head” is not all that
serious. A closer look shows that that is not
the case.
When patients are tested using laboratory
tasks evaluating working memory functions,
effects will be found. However, the labora¬
tory tests are considerably more searching
than the memory evaluations included in
an intelligence test. The people who have
been injured often do not report problems
in everyday life, even though they have trou¬
ble with the laboratory tests. We might then
conclude that the residual effects of concus¬
sion are not serious enough to be of concern.
But that is not the case.
Investigators at the United Kingdom’s
Applied Psychology Unit (Cambridge) went
a step further. They asked the living part¬
ners of people who had suffered concus¬
sions whether or not there had been any
long-term effects. The partners replied that
there had been, and furthermore, the sever¬
ity of the deficit, as reported by the partner
but not the affected individual, was corre¬
lated with the difficulty the affected person
had in dealing with the laboratory tasks.30
In addition to being interesting in itself, this
study showed the importance of obtaining
information about a person’s everyday per¬
formance, as well as observing performance
29 Information downloaded from www. healthline.
com/adamcontent/concussion, June 2009.
30 Sunderland, Harris, & Baddeley, 1983.
during an out-of-context situation, such as
a laboratory study or a conventional testing
session.
A second study shows how concussion
can act as a distal influence that increases
the risk of incurring a condition that acts as a
proximal influence on intelligence. The peo¬
ple studied were elderly (sixty-plus) pairs of
twins, where one twin suffered from Parkin¬
son’s disease, which results in a loss of intel¬
ligence in its latter stages, and the other did
not. The risk of incurring Parkinson’s dis¬
ease tripled for those who had experienced
head injuries, even though the head injuries
had occurred, on average, thirty-seven years
before the study was conducted.31
As these two studies illustrate, long-term
damage to intelligence can result from what
are, at the time, apparently recoverable
injuries to the brain. Stories of punch-drunk
boxers are not just legend.32
9.3.2. Prenatal and Infant Health Issues
A great deal of attention has been devoted
to prenatal and infant development, because
this is when the foundations of cognition are
laid down. In their first year infants acquire
a great deal of information about their lan¬
guage and about social interactions.33 To
what extent do individual differences in
cognitive function in infancy predict later
indices of intelligence? Infancy is also a
period of high physiological vulnerability.
To what extent are these early indicators
sensitive to environmental variables?
Infant development is often measured by
the Bayley scales of development, devel¬
oped originally by Nancy Bayley and her
colleagues at the University of California,
Berkeley, and updated periodically.34 These
scales document the age of normal occur¬
rence of a variety of behaviors, such as
31 Goldman et al., 2006.
32 Nor do we need to rely on legend. Survey results
have shown that the incidence of memory prob¬
lems is markedly elevated among retired profes¬
sional football players {New York Times , Sept. 30,
2009, p. Ai].
33 Gopnick, Meltzoff, & Kuhl, 1999.
34 Bayley, 2005.
2?2
HUMAN INTELLIGENCE
crawling, toddling, vocalization, and reac¬
tivity. Contemporary middle- and upper-
class parents are notoriously concerned that
their children be on schedule. This con¬
cern may be a bit overdone. The correlation
between scores on developmental scales that
are based on activity and vocalization over
the first thirty-six months of life, and scores
on adult intelligence tests is nearly zero.35
However, this does not mean that the devel¬
opment of information-processing and ver¬
bal measures can be disregarded. By the age
of three, tests that involve vocabulary cor¬
relate in the q-q range with IQ scores for
young adults.
One of the most important things an
infant has to do is to recognize stability
and change in the environment. The infant’s
ability to do this can be evaluated by mea¬
suring habituation, the ability to discrimi¬
nate between novel and repeated stimuli.36
Let us take a closer look at the procedure
used.
Joseph Fagan, a professor at Case Univer¬
sity, and his colleagues have measured habit¬
uation by showing infants an "interesting”
picture, often a picture of another infant.
After the infant has viewed one picture it
is removed, and then presented again, along
with a new picture. The measure of habitu¬
ation is the extent to which the infant looks
at the new picture. This simple test for six-
to twelve-month-old infants has a correla¬
tion of .32 (corrected for reliability, .59) with
measures of adult intelligence at twenty-one
years of age.37
Birth weight, one of the most commonly
used indices of prenatal health, is corre¬
lated with intelligence, for both premature
and normal term infants. One study esti¬
mated that there is an increase of between
.3 to .5 IQ points for every 100 grams of
birth weight over 2,500 grams (about 5.5
pounds).38 This finding, standing alone, is
ambiguous, because maternal IQ is posi-
35 Bayley, 1968.
36 Brody, 1992, p. 232; Fagan, Holland, & Wheeler,
2007.
37 Fagan, Holland, & Wheeler, 2007. See also the ref¬
erences therein for citations to earlier studies.
38 Matte et al., 2001.
tively correlated with the birth weight of
the child. Collinearity again! When maternal
IQ is controlled statistically, the relationship
between IQ and birth weight drops but does
not disappear.39 The finding is not confined
to cases of abnormally low birth weight.
Differences in intelligence between heav¬
ier and lighter babies in the normal range
(2,000 grams and above) have been reported
in studies in a variety of North American
and European countries, and in both white
and African American populations in the
United States.40
The relationship does not solely reflect
any tendency for bright mothers to have
children that are both heavier and have a
higher genetic potential for intelligence, as
separate effects. This was shown in an ele¬
gant study of two large samples, one in New
Zealand and one in the United Kingdom.41
Because of the size of the samples it was pos¬
sible to compare birth weights and IQ scores
in monozygotic (MZ) twins, thus controlling
for genetic influences. The heavier twins had
higher IQ scores, demonstrating the impor¬
tance of the prenatal environment indepen¬
dent of genotype.
We can take the comparison to an
extreme by looking at “preemies,” new¬
borns with a gestational age of less than
thirty- two weeks, many of whom have birth
weights well below 2,000 grams. These chil¬
dren, on the average, have IQ scores almost
one standard deviation unit below a con¬
trol group.42 In general, nonverbal function¬
ing and abstract reasoning are more affected
than verbal functioning. Considering the
links between information processing and
g, it is not surprising to find that prema¬
ture infants do poorly in childhood (eight
to twelve years) on tests of working mem¬
ory and attentional control.43
Because birth weight predicts future
intelligence, it is of interest to know what
variables may lead to low birth weight. By
far the largest risk factor is premature birth,
39 Deary, Der, & Shenkin, 2005.
40 Dombrowski, Noonan, & Martin, 2007.
41 Newcomb e et al., 2007.
42 Esbjorn et al., 2006.
43 Bayless & Stevenson, 2007.
ENVIRONMENTAL EFFECTS ON INTELLIGENCE
273
Panel 9.4. Fetal Alcohol Syndrome
Fetal alcohol syndrome (FAS) is a defect
in intelligence occuring in children whose
mothers drank heavily (more than four
drinks per occasion) during pregnancy.
Prevalence in the US is somewhat less
than one case per thousand births.
Prevalence is higher in social groups
that have high rates of alcohol abuse.
The problem is particularly acute in the
Native American population, where the
incidence of FAS has been estimated to
be from 1.5 to 2.5 cases per thousand
births.
The effects can be devastating. FAS
is characterized by facial malformation
(short noses, small eye openings, thin
upper lips, and skin folds over the eyes,
among other features) and severe cogni¬
tive retardation. A broader term, fetal
alcohol spectrum disorder (FASD), is
used to refer to prenatal alcohol damage
that includes all damage due to alcohol
consumption during pregnancy.
The National Institute of Health
guidelines advocate complete abstinence
during pregnancy. Damage can be
inflicted early in the first trimester of
pregnancy, so a woman might damage the
fetus before becoming aware that she is
pregnant.
The fetus can be put at risk by drink¬
ing in the social range, below the level
of intoxication. Anne Streissguth and her
group at the University of Washington
Medical School have conducted longitu¬
dinal studies of the children of women
who either abstained completely or were
social drinkers during their pregnancies.
Among the latter, small but clear indi¬
cations of impulsive decision making and
difficulties with attentional control were
detected in their children when they were
fourteen years old. The degree of impul-
sivity varied with the amount drunk dur¬
ing pregnancy, even though none of the
mothers would be considered to have
been alcoholics.*
General information on fetal alcohol syndrome was
downloaded from wwTv.nlm.nih.gov/medlinepius/
fetalalcoholsyndrome.html#cati, June 2008, and
links derived from this page.
* Sampson et al., 1997; Connor et al.; 2001.
although a myriad of other causes can also
affect maternal and fetal health. The opti¬
mal age for childbearing, at least as far as
the risk of low birth weight is concerned, is
from twenty-five to thirty-two; women out¬
side this age range are at heightened risk for
premature birth and for having a low birth
weight child.
Not surprisingly, expectant mothers are
strongly advised to avoid ingesting any sub¬
stance that is toxic to neural development.
In our own society perhaps the common¬
est such substance is alcohol. Social drink¬
ing can have an effect, and maternal drink¬
ing to the point of serious intoxication can
have devastating influences on fetal intellec¬
tual development. This is discussed in more
detail in panel 9.4.
We seem to be close to the situation
we encountered with respect to molecular
genetics; we know neonatal-infant environ¬
mental effects that can harm intelligence.
Is there any way to improve intelligence,
within the normal range? Some interesting
things have been tried, but there is little
objective evidence that various diets, exer¬
cise regimens, or even having the expec¬
tant mother listen to classical music (see
panel 9.5) does very much good. The way
to make sure children are born highly intel¬
ligent still eludes us.
9.3.3. Nutrition
Nutrition , in the broadest sense, refers to
anything that people ingest as food or drink.
This excludes certain substances that are
ingested but not for the purpose of fuel¬
ing metabolism, such as the ingestion of a
psychoactive drug. There are a few cases of
274
HUMAN INTELLIGENCE
Panel 9.5. The Mozart Effect
There has been a good deal of inter¬
est in the possibility that listening to
music - and for some reason, listening
to the works of Mozart - might improve
spatial-visual reasoning.* The claim has
even been made that the effect can be
observed in rats! The phenomenon has
been dubbed the "Mozart effect."
The first reports of the effect led
to a crescendo of replications, and an
exchange of letters on the topic in Nature
in 1999. 1 My own belief is that the effects,
if any, are small and transient, and are
probably due to structured music tem¬
porarily stimulating some of the brain
regions used to solve spatial-visual prob¬
lems. It seems unlikely but not impos¬
sible that a permanent enhancement of
spatial-visual reasoning results. Edward
Zigler, a highly respected specialist in
developmental psychology, has criticized
the fascination with the (alleged) effect
as yet another search for a "quick fix” in
early childhood education and support.
Zigler worries that the attention focused
on such fads distracts public attention
from the development of less dramatic,
more expensive, high-quality infant care
and pre-school programs.4
* Rauscher, Shaw, & Ky, 1993.
Chabris, 1999.
1 Jones & Zigler, 2002.
substances that have been ingested for nutri¬
tional purposes, but that contain pathogens
that can produce a loss of intelligence.
Panel 9.6 describes two such cases.
Intuitively, it makes good sense to think
that nutrition will influence intelligence,
especially during periods of neural growth.
It turns out to be difficult to prove this,
for several reasons. We want to distin¬
guish between temporary effects, during
a period of malnutrition, and permanent
effects, exhibited following recovery from
malnutrition. It is also necessary to distin¬
guish between the effects of brief periods
of malnutrition (and when, in development,
these periods take place) and the effects of
chronic malnutrition. The type of malnutri¬
tion is also an issue. A diet may have ade¬
quate caloric intake and still be deficient in
protein, iron, or other substances important
for neural development. There is also the
problem of distinguishing between cogni¬
tive and temporary attentional effects. Mal¬
nourished people are physically weak and
have trouble concentrating. This may lead
to underperformance both in a testing ses¬
sion and, more importantly, in any situa¬
tion involving cognitive demands over a long
period of time.
These are conceptual problems. There
are also some major practical problems.
It would not be ethical to subject peo¬
ple to malnutrition in a controlled study,
so correlational and epidemiologic studies
are necessary. Reliable records of nutritional
intake may not be available. Collinearity is
an issue. Malnutrition to a level that would
influence neural development, and hence
intelligence, is rare in the industrially devel¬
oped countries. When it does occur it is
likely to be accompanied by poor general
parenting practices and a lack of social sup¬
port for mother and child. In the develop¬
ing countries malnutrition does occur, but
it tends to occur in the poorest areas of
the country, those with the least access to
schools and medical facilities. In general,
more intelligent, better-educated mothers
provide better nutritional environments for
their children.44 Genetic and social effects
on the child's intelligence can appear to be
due to nutrition unless care is taken to mea¬
sure the appropriate covariates.
Nonetheless, the problem is an impor¬
tant one, so researchers have persevered. We
44 For example, a study of Egyptian mothers by W achs
& McCabe, 2001.
ENVIRONMENTAL EFFECTS ON INTELLIGENCE
275
Panel 9.6. Two Cases of Very Bad
Dietary Practices
In America and Europe today a great
deal of publicity is given to dietary prac¬
tices. Most of that concern is directed
toward obesity and cardiovascular health.
There are some cases, though, in which
the concern is overingestion of foods con¬
taining pathogens that may influence the
brain. The famous Salem Witch Trials in
seventeenth-century Massachusetts were
a response to bizarre behavior by chil¬
dren. A case has been made that the
behavior may have been due to a neu¬
rotoxic fungus in bread that the chil¬
dren ate.* But this is only speculation.
Here are two better-documented exam¬
ples. One involves an exotic social prac¬
tice that, thankfully, is no longer with us.
The other illustrates the dangers that can
result when a pathogen gets into the vast,
almost unseen food chain in industrially
developed nations.
Kuru is a disease that attacks the brain
and central nervous system. It is caused
by a slowly acting agent called a prion that
destroys neural structures/ In the 1950-
60 period kuru broke out in the high¬
lands of New Guinea, then one of the
world’s most primitive areas. The out¬
break was due to cannibalism; people
became infected by eating the brains of
people who were already infected with
kuru. The connection was not immedi¬
ately obvious because of the long period
of time that typically elapses between
infection and the manifestation of the
disease. Today kuru has virtually disap¬
peared because of the modern Papua
New Guinea government’s policy of
strongly discouraging the practice of can¬
nibalism/
A related brain disorder, Creutzfeldt-
Jakob disease, can be acquired by inges¬
tion of infected tissue from meat. (The
risk of acquiring the disease is also
partially under genetic influence.} The
purity of meat is regulated by social prac¬
tices, especially the screening of cattle
for the presence of a bovine form of
the disease (“mad cow” disease] prior to
slaughter and rendering into commercial
beef/ The fact that infected meat could
produce Creutzfeldt-Jakob disease was
first discovered in Britain in 1996, fol¬
lowing a sharp rise in the incidence of
the disease. Since then screening of cat¬
tle has greatly increased. Nevertheless, in
2008 there were major political protests
when the government of the Republic of
Korea decided to permit importation of
US beef although mad cow disease had
been detected in the US herd.
* Caporael, 1976.
* Gajdusek & Zivas, 1957. Gajdusek received the
Nobel Prize in Medicine in 1976 for the discovery
of slow-acting prions.
1 Information recovered from www.ninds.nih.gov/
disorders/kuru/kuru.htm, June 2008.
$ Information recovered from www.ninds.nih.gov/
disorders/cjd/detaiI_qd.htrn, June 2008.
will look at a few key studies. Most of these
have involved infants or young children, as
they are perceived to be most vulnerable to
the effects of malnutrition. There are a few
studies of the relation between nutrition and
adult cognition.
Brief periods of malnutrition do not
appear to leave permanent effects, even
when the malnutrition occurs in utero or
in infancy. The evidence for this comes
from a large, quasi-experimental study that
occurred in the Netherlands as a byprod¬
uct of the battles in Western Europe dur¬
ing World War II. The events and the study
are described in panel 9.7. The gist of the
findings was that male children exposed to
a few months of intense starvation during
their infancy did not show any effect of that
experience on cognition, when they were
tested nineteen years later. Apparently the
effects of severe short-term malnutrition are
reversible.
276
HUMAN INTELLIGENCE
Panel 9.7. The Hunger Winter Study
A dramatic and tragic incident in World
War II provided one of the best demon¬
strations that temporary malnourishment
does not have a permanent effect on cog¬
nition.
At the beginning of 1944 Western
Europe was occupied by Nazi German
forces. In June 1944 American and British
troops, together with smaller Allied con¬
tingents, landed in France and began
a steady march eastward toward Ger¬
many. In September 1944 British para¬
troopers attempted to seize bridges on
the Rhine at Arnhem in the Netherlands.
The plan was for the paratroopers to link
up with British ground forces advanc¬
ing from the southwest. The combined
forces would then cross the bridges into
Germany.
The Dutch railway workers went on
strike, hoping to cut off supplies to the
Germans. However, the British ground
forces were unable to advance, and the
paratroopers were overrun. To retaliate
against the Dutch the Nazis imposed a
transportation embargo on the northern
Dutch cities under their control. Star¬
vation ensued. At one point the esti¬
mated caloric content of rations was
down to 640 calories per adult per day.
(Depending on age and size, the recom¬
mended US daily caloric intake is some¬
what over 2,000 calories.) Meanwhile the
southern Dutch cities, under Allied con¬
trol, received adequate food. The starva¬
tion period lasted from September until
March, when American troops seized
a bridge across the Rhine at Remagen.
Allied forces crossed rapidly into Ger¬
many, and the Dutch cities were relieved
and supplied. The war in Europe ended
shortly thereafter.
In the Netherlands men register for
military service at age nineteen. Reg¬
istrants are given a progressive matrix
test. Zena Stein and her colleagues
from Columbia University compared the
scores of men who were either in utero
or neonates during the four months of
the starvation period to the scores of
men who had been born at the same
time, but in the cities under Allied con¬
trol. Other records of mental capabilities,
such as the frequency of mental retarda¬
tion, were also compared across the two
populations. There was no difference in
any indicator of mental competence. This
study of over 125,000 men provides strik¬
ing evidence of the resilience of human
cognition.*
* Stein et ah, 1972.
Chronic, prolonged malnutrition is some¬
thing else. Numerous studies in develop¬
ing nations have provided evidence that
prolonged malnutrition is associated with
low test scores, particularly on nonverbal
tests. A closer look reveals an interesting
pattern of defects. Malnourished children
are described as less attentive, impulsive,
and easily distracted. Researchers primar¬
ily interested in nutrition describe this as
a confounding variable, saying that it is
unclear whether the effects of malnutri¬
tion are on intelligence, or whether they
are on attention.45 If we consider these
45 Sigman & Whaley, 1998.
observations in the light of studies of the
role of basic information processing in intel¬
ligence, we reach a somewhat different con¬
clusion.
The behaviors that characterize malnour¬
ished children are indications of lack of
attentional control, which is a vital part of
the working memory complex. Individual
differences in working memory and the con¬
trol of attention are highly related to intelli¬
gence, especially measures of g and Gf. Pro¬
longed malnutrition also influences habitua¬
tion, which, as we have seen, is an indicator
of infant intelligence.
Viewed in this light, malnutrition directly
influences intelligence test scores, because
ENVIRONMENTAL EFFECTS ON INTELLIGENCE
the act of taking the test requires exer¬
cise of the working memory complex. The
same deficiencies in information process¬
ing influence young children's school per¬
formance. This produces a deficiency in the
knowledge-based aspects of intelligence. A
physical agent, malnutrition, results in an
inability to benefit from environmental sup¬
ports that improve cognition.
This conclusion is reinforced by stud¬
ies of the beneficial effects of interven¬
tion. Two sources of intervention have been
tried: improvements in general nutrition and
improvements in intake of specific nutrients,
primarily iron and protein. Iron is important
because iron-deficiency anemia leads to gen¬
eral listlessness and, once again, inability to
focus attention. Proteins are considered nec¬
essary for development of the neural system.
The results of a study conducted in rural
Guatemala are particularly informative. Pro¬
tein supplementation was provided for some
children, while a nonprotein supplement
was offered to others. Protein supplemen¬
tation improved various measures of cog¬
nitive performance. In addition, there was
an important interaction. The greatest gains
occurred when protein supplementation
was combined with school attendance.46
This is consistent with the argument that
appropriate nutrition will benefit working
memory and related attentional processes.
These then make it possible to take advan¬
tage of an environment that supports the
development of cognitive skills. For optimal
effect a program to help children recover
from malnutrition should be combined with
an educational program.
The sorts of serious malnutrition that lead
to cognitive deficits are generally not found
in the industrially developed countries. In
these countries excess caloric input, which
can lead to a quite different set of problems,
is more of an issue. When severe malnutri¬
tion occurs in the developing countries, it
is usually because of a situation over which
the individual family has little control - war,
famine, or drought. The next physical envi¬
ronmental effect that we look at, alcoholism,
46 Pollitt et al., 1993.
277
is a matter of lifestyle, and is found widely
in Europe, the Americas, and in industrial
Asia.
9.3.4. Alcohol Abuse
Alcohol is the commonest, and most abused,
recreational drug in the world. We have
already discussed the effects of mater¬
nal alcohol consumption on the fetus
(panel 9.5). Here we concentrate on the
effects on the consuming adult.
While definitions of abuse have shifted
over time and place, present psychiatric cri¬
teria distinguish two forms of alcoholism:
alcohol dependence, in which the affected
person can scarcely go without a drink, and
alcohol abuse, which refers to people who
consume alcohol frequently, and heavily,
but can go for periods of time without drink¬
ing. According to the US National Institutes
of Health (NIH) statistics for 2001-02, 75
of every 1,000 adults over the age of eigh¬
teen, or 7.5%, fell into one of these two
categories. (The figure includes "recovering
alcoholics,” who are not currently drinking.]
For reference, the incidence of heart dis¬
ease, the commonest cause of death in the
US, is slightly greater than 8%. There is a
genetic component to alcoholism; the close
relatives of alcoholics are at heightened risk
for alcoholism.4 However, the genetic com¬
ponent is by no means a determiner.
Alcoholism is diagnosed using four
behavioral criteria. They are:
Craving: A strong compulsion to drink.
This goes well beyond thinking it
would be nice to have wine with your
meal.
Loss of control: The inability to cease
drinking once drinking is begun.
Withdrawl symptoms: Nausea, sweating,
shakiness, and anxiety when alcohol
use is stopped.
Tolerance: The need to drink ever greater
amounts of alcohol in order to
experience symptoms of intoxication,
including relaxation and euphoria.
47 McGue, 1999.
278
HUMAN INTELLIGENCE
Brain imaging studies of alcohol abusers,
compared to age-matched controls, have
shown something that had long been sus¬
pected on behavioral grounds.48 Prolonged
alcoholism damages the frontal lobes 49 This
is accompanied by decrements in abstract
reasoning and in visual-spatial reasoning,
the sorts of behaviors that one associates
with decreased working memory, loss of
the ability to plan actions, and diminished
attentional control. 5° Extremely heavy, pro¬
longed drinking can produce damage to the
limbic system, with an associated loss of the
ability to store new memories (. Korsakoff's
syndrome). Korsakoff’s syndrome patients
are unable to function in society, and hence
must be hospitalized.
The results are clearly due to alcohol
abuse, and not to pre-existing genetic or
familial conditions, because the effects of
excess use of alcohol can be demonstrated
in twins, where one twin is an abuser and
the other is not.51
The issue with regard to social drinking
is less clear. Surveys of social drinkers indi¬
cate that there is a small negative associa¬
tion between moderate social drinking and
cognitive test scores. Not surprisingly, the
effect is most marked among those who con¬
sume alcohol more than four times a week,
and who regularly consume more than 40-50
ml of alcohol per occasion. This is roughly
equivalent to three or four drinks of hard
liquor, glasses of wine, or bottles of beer.
There are marked individual differences in
tolerance. The critical amount varies with
weight and with sex, women generally being
more sensitive. In terms of psychometric
models, the effects appear to be on g or
Gf, depending upon whether the g-VPR or
three-stratum model is used to describe the
48 There are many studies showing deficiencies in
reasoning and intelligence associated with alcohol
abuse. Clarke & Haughton, 1975, and Leckliter &
Matarazzo, 1989, are typical examples.
49 Moshelhy, Georgiou, & Kahn, 2001.
50 See Schottenbauer et al., 2007, for a recent example
of such a study and note how these results, using
imaging technology, expand but do not alter the
conclusions of a study by Jones (1971) thirty-five
years earlier.
51 Toomey et al., 2003.
results. There is not enough data to make
a clear statement about the effects of social
drinking upon visual-perceptual reasoning.52
Collinearity makes it hard to define
causality. The extent of social drinking
varies greatly among different racial, ethnic,
and demographic groups. In some circles
social drinking is an accepted and (almost)
expected practice. In others any use of alco¬
hol is frowned upon. It is difficult to dis¬
entangle the effects of social drinking from
inherited intellectual potential, health prac¬
tices, and other lifestyle variables. There is
also a chicken-and-egg problem: does heavy
social drinking reduce intelligence, or is it
the case that the intelligent person does not
drink heavily?
This is hard to say, because there is data
indicating that childhood intelligence tests
(i.e., measures of cognitive power taken
before people begin to drink) predict drink¬
ing patterns. People with higher test scores
are less likely to drink to the point of hav¬
ing a hangover, surely an intelligent thing to
do, and are more likely to consume wine,
which is generally drunk more slowly and
more likely to be drunk with meals, than
whiskey or beer.55 This distinction is impor¬
tant, because the toxic effects of alcohol are
related to a buildup of alcohol metabolites
in the bloodstream. This occurs when alco¬
hol is taken in at a faster rate than it can be
processed through the liver. When alcohol is
taken with food, the rate of absorption from
the stomach to the bloodstream is reduced,
thus lessening the influence of the drug.
While admitting that the issue is not
clear, the evidence favors the hypothesis
that repeated heavy drinking, to the point of
feeling somewhat “high,” although not nec¬
essarily to the point of losing consciousness
or marked motor control, does lead to cog¬
nitive deficit.
The effects of alcohol on intelligence are
extremely important on a population basis.
According to a Center for Disease Control
52 See Parker, Parker, & Harford, 1991, for a review and
discussion of this research.
55 Batty, Deary, & McIntyre, 2006; Mortensen,
Sorensen, & Gronbaek, 2005.
ENVIRONMENTAL EFFECTS ON INTELLIGENCE
279
survey taken in 2006, slightly more than 60%
of US adults twelve or over described them¬
selves as current drinkers. About 22% of the
adults surveyed reported at least one inci¬
dent of binge drinking, which was defined
as five drinks or more per occasion. This
is well over the danger level suggested by
the literature. Binge drinking was heaviest
among people eighteen to twenty-five (43%
of the group), and within this group heaviest
among college students. The famed (or infa¬
mous) weekend fraternity bash is the most
dangerous mode of consumption a social
drinker can engage in.
We now turn our attention from what we
put into our mouths to what we breathe in
from the air.
9.3.5. Atmospheric Lead
Modern environmentalists worry a great
deal about industrial pollutants in the air,
water, and soil. The problem is not new.
Mercury nitrate, a neurotoxin, was used in
hat making until the twentieth century. A
century earlier the phrase “mad as a hatter’
was in common use, for hat makers were
thought to have delusions and poor motor
control. Today mercury’s dangers are well
known, and industrial exposure is tightly
regulated. There is considerable concern
over the presence of trace amounts of mer¬
cury in the food chain, especially in cer¬
tain fish. A much commoner pollutant, lead,
presents a larger problem, worldwide.
Lead was one of the first metals to be
mined, for reasons ranging from its mal¬
leability to its taste (I).54 Analyses of human
remains show that prior to the beginning of
metallurgy, about 4000 BCE, the concentra¬
tion of lead in the human body was .0016
micrograms per deciliter (pg/dL). In 1975—
80 the concentration in American children
was estimated to be 15 pg/dL, over 1,000
times the natural level. The concentration
has fallen markedly since then, due largely
to the banning of tetraethyl (“lead added”)
gasoline.55
54 The Romans used lead sulfate to sweeten wines.
55 Hubbs-Tait et al., 2005.
The controversy over, and eventual ban¬
ning of, lead additives in gasoline did not
represent a newfound concern over lead.
Debates over the costs and benefits of using
the metal can be traced back to the time of
the Roman Empire. Panel 9.8 presents some
of the history. It shows the difficulty of strik¬
ing a balance between the undeniable eco¬
nomic benefits of an activity and the equally
undeniable cost of risks to public health.
The earlier controversies were over the
effects of the levels of exposure that might
be experienced in an industrial operation
using lead. Today's concerns are over the
cumulative effects of exposures to much
lower concentrations of atmospheric lead,
and in particular the effects upon children.
These concerns were heightened by results
reported by Herbert Needleman of the Uni¬
versity of Pittsburgh in 1979. 56 Needleman
realized that a large-scale epidemiological
survey of lead concentrations in children's
bodies could be conducted, in a highly
noninvasive manner, by collecting young
schoolchildren’s “baby teeth” after they fell
out and analyzing them for lead content. He
then contrasted the Wechsler test (WISC-
R) performance of first and second grade
children with high and low concentrations
of lead in their teeth. Higher concentra¬
tions were associated with low IQ scores.
In a second study the level of lead in the
body was positive correlated with teach¬
ers’ reports of children’s impulsive behav¬
ior. The researchers allowed for the effects
of a number of control variables, including
parental SES. On its face this was an impres¬
sive study of over 2,000 children.
Needleman made the dramatic claim that
lead concentrations could be responsible for
a six- to seven-point IQ drop in children.
Given the substantial economic implica¬
tions of this assertion, it is hardly surpris¬
ing that his work was challenged. Some of
these challenges amounted to accusations
of improper manipulation of data. These
charges were investigated, and Needle¬
man was exonerated of any wrongdoing.
56 Needleman et al., 1979.
280
HUMAN INTELLIGENCE
Panel 9.8. The History of Concerns
over Lead Poisoning
Lead has been used since antiquity. Its
widespread use was accompanied by
unheeded warnings.
The Romans used lead in dishes and
in water pipes, in spite of the architec¬
tural engineer Vitruvius’s warning that
lead could ‘"rob the limbs of the virtues of
the blood.” Seventeen hundred years later
Benjamin Franklin gave a similar warning.
Franklin, a shrewd observer of both phys¬
ical and social realities, said that his warn¬
ing would be ignored because of the eco¬
nomic benefits of using the metal. In the
early twentieth century Alice Hamilton,
M.D., the first woman admitted to the
Harvard faculty, documented over 500
cases of lead poisoning. She observed that
her findings were consistent with those of
French authorities one hundred years ear¬
lier. The public health problem became
acute with the advent of the automobile.
Tetraethyl lead in gasoline is a cost-
effective anti-knock agent. The first
refineries for leaded gasoline were devel¬
oped after World War I by a cooperative
effort of General Motors, Standard Oil
Company (now Exxon), and DuPont. In
1924 there were outbreaks of insanity, ill¬
ness, and death in two of the new refiner¬
ies. Various controversies, investigations,
and court cases occurred from the 1920s
through the 1940s. The automobile indus¬
try avoided being regulated, although
the evidence against lead mounted and
mounted. The debate was bitter. Alice
Hamilton personally confronted Charles
Kettering, General Motors vice president
for research (and a famous automotive
engineer), and called him a murderer!
In the 1970s legislation was passed to
remove lead from gasoline, due to the
perceived public health hazards. This
decision cost the automotive industry bil¬
lions of dollars, a cost that was promptly
passed along throughout the economy.
This did not clear the air, for the lead
produced by automobile combustion is
only part of the problem. Many industrial
processes produce atmospheric lead as a
by-product. Removal costs can run into
millions of dollars. Therefore, there was
further debate over whether other pro¬
cesses, including lead smelting, produced
a sufficient health hazard to justify the
economic costs of controlling emissions.
The consensus now is that they do, and
in the developed industrial countries the
emission of lead into the atmosphere and
the use of lead in home products are both
strictly regulated.
Nevertheless, many products used in
the home are sources of lead. These
include paints (especially older versions,
still found in many homes), polyethylene
plastic bags, and even candy wrappers. In
November of 2007 some Chinese-made
toys intended for the US Christmas mar¬
ket were found to contain unacceptable
levels of lead. Senator (later President)
Barack Obama, then involved in a tight
race for the Democratic nomination for
president, made the sweeping statement
that if elected he would bar the impor¬
tation of toys from China. Obama subse¬
quently retreated from this promise, pos¬
sibly because 80% of US toys are made in
China.
The laws regulating lead present costs
to industry that may reach into the
billions of dollars. On the other side of
the coin, the CDC has estimated that
in the early 2000s 1.1% of US children
under the age of five had blood lead
concentrations higher than the current
allowable level of lopl/dl. As of 2007, the
Census Bureau estimated that there were
20.75 million children in this age bracket.
That comes to 228,250 children who may
be building up dangerous concentrations
of lead. This is a significant health risk.
The problem is confounded by the fact
ENVIRONMENTAL EFFECTS ON INTELLIGENCE
281
that lead sources tend to be located in
poorer residential areas, close to indus¬
trial operations.
It would be impossible, and inap¬
propriate, for a book on intelligence to
attempt to weigh the relative costs and
benefits of lead use, or to comment on the
costs and benefits of similar industrial-
public health trade-offs. What we can do
is examine the research that has caused
us to be concerned.
The information in this summary comes from
Hubbs-Tait et ah, 2005, and Kovarik, 2005.
Collinearity posed a more difficult problem
in interpretation. Was the poor test perfor¬
mance of children with higher lead con¬
centrations caused by the concentration of
lead, or did they perform poorly because
they tended to come from poorer homes
and/or to be minority group members, and
therefore would be expected, on statistical
grounds, to have lower scores on intelligence
tests? The best way to investigate this issue is
by replication of a result, in situations where
the collinearity problems plaguing the orig¬
inal finding do not occur.
Several such replications have been con¬
ducted. One is a longitudinal study of chil¬
dren living near industrial operations in
Port Pirie, Australia.57 A second longitudi¬
nal study following up children from two
to seven was carried out in Kosovo in the
1990s.58 In both these studies investigators
obtained maternal IQ scores and took exten¬
sive measures of the home environment. In
the Kosovo study two separate populations
were studied, children living in a relatively
large town near a lead smelter and chil¬
dren in a similar town, twenty-five miles
away, where the atmospheric lead expo¬
sure level was much lower. Pregnant moth¬
ers were enrolled, prenatal lead concentra¬
tions were estimated, and children were fol¬
lowed up until they were seven in order
to assess the effects of a buildup of lead
in the body. Both prenatal levels of lead
and postnatal increases were associated with
drops of intelligence after all other vari¬
ables were considered. There was no evi-
57 Baghurst et al.f 1992.
58 Wasserman et al.f 2000. The study was carried out
at a time when Kosovo was part of Serbia, and was
not disrupted by war.
dence of threshold effects, which provides
an argument against maintaining a permissi¬
ble concentration level of 10 pm/dl, as is now
done.
Lead concentrations were statistically
associated with just over 4% of the variance
in children’s IQ test scores. Similar results
have been obtained in other studies in the
US and in South America, so we have a
highly generalizable finding. 59
In 2005 a consortium of researchers
involved in these studies published an analy¬
sis of the international findings.60 They con¬
cluded that there is a non-linear relationship
between intelligence test [IQ] scores and the
level of lead in the blood. Their guide was
Increment in lead
Expected loss
in blood
in IQ points
2.4 to 10 yd dL
3-9
10 to 20 p/dL
1.9
20 to 30 yddL
1.1.
The losses are cumulative, so the expected
loss for a child with a 30 yd dL concentration
of lead in the blood would be 6.9 IQ points,
Needleman’s original estimate.
A drop of five to seven IQ points is not
serious on an individual basis. We would not
expect to find a great deal of difference in
the cognitive capabilities of two children,
one with an IQ of 100 and the other with an
IQ of 95. On a population basis, though, this
would be a serious issue, because of changes
59 Wasserman et al., 2000, p. 815. See related work in
the United States by Chiodo et al. (2007) and in
South America by Counter, Buchanan, & Ortega
(2005).
60 Lanphear et al., 2005.
282
HUMAN INTELLIGENCE
Table 9.1. Levels of blood lead concentration in children, together with recommended
actions
Class
Blood Lead
Concentration
( V-g/dl)
Comment
I
Less than 10
Not lead poisoned.
IIA
10-14
Children in Class IIA should be screened frequently. If many
children in a community test in this range, community-wide
preventive measures should be taken.
IIB
15-19
A child in Class IIB should receive nutritional and educational
interventions and more frequent screening. Environmental
investigations and interventions should be initiated.
III
20-44
A child in Class III should receive a medical evaluation. The child
may require pharmacological treatment for lead poisoning.
Environmental evaluation and remediation is called for.
IV
45-69
A child in Class IV requires medical and environmental
intervention, including chelation therapy (a technique for removing
lead from the body).
V
70 or above
The child is suffering from lead poisoning. This is a medical
emergency. Medical intervention and environmental management
must begin immediately.
Note: Information provided by the US Center for Disease Control brochure “Preventing Lead
Poisoning in Young Children/’ published in October 1991. The commentary has been paraphrased
from comments in the brochure.
in the frequency of exceptionally high and
low levels of intelligence. Consider a pop¬
ulation of a thousand children. If the mean
IQ in the population were ioo; we would
expect to find approximately fifty children
with IQs below 70, which is often consid¬
ered an indication for enrollment in a special
education program, and an equal number of
children with IQs above 130, in some pro¬
grams a marker for entry into a gifted educa¬
tion curriculum. If the mean of the popula¬
tion dropped to 95, we would expect to find
about one hundred children with IQs below
70, and only twenty-five with IQs above 130.
In other words, a drop of five IQ points in
the population average would be associated
with a doubling of the number of children
in the special education program, while the
number of children eligible for the gifted
program would drop by a half.
These research findings have had an effect
on public policy, at least in the industrial and
postindustrial countries. In i960 the border
between safe and unsafe exposure levels was
60 p/ dL, twice the level used as a point of
concern in the 2005 comprehensive review.
Table 9.1 shows the current guidelines from
the US Center for Disease Control (CDC).
Actions to reduce lead concentrations are
now recommended when children’s blood
levels exceed 10 p/dL, and a blood level
above 45 p/dL is seen as a medical emer¬
gency. An advisory committee report pub¬
licized by the CDC has indicated that con¬
centrations higher than 10 p/dL can warrant
actions to reduce the levels of lead in the
home or near schools.^1
Unfortunately, significantly higher levels
of atmospheric lead are found in some devel¬
oping countries. For instance, in the indus¬
trial port of Callao, near Lima, Peru, lead
storage areas are located near one of the
61
CDC Advisory Committee, 2007.
ENVIRONMENTAL EFFECTS ON INTELLIGENCE
283
poorest residential areas of the city. A survey
found that approximately half of the early
elementary school children had blood lead
levels of 20 \U dL or higher.62
Lead in the atmosphere clearly represents
one of the major threats to the development
of intelligence in some areas of the world
today.
9.4. The Social Environment
The social environment is an important
determiner of cognitive power. Charles
Murray has called the development of cer¬
tain ways of thinking “meta-inventions/'
because societies and individuals who use
these ways of thinking have a huge advan¬
tage over those who do not.6* Literacy
is probably the most important of the
meta inventions, because it fosters abstract
thinking and provides continuity with the
past. Mathematical reasoning and scientific
approaches to problem solving also improve
thinking. In our society the intelligent per¬
son is one who has a good grasp of these
tools of thought.
Americans spend a good deal of time and
effort trying to ensure that their children
become familiar with the tools of intelli¬
gence. The effort literally begins at home.
The Disney Corporation’s Baby Einstein
DVD programs for three-month- to three-
year-olds are supposed to enhance intelli¬
gence in toddlers. Whether they do so is
questionable.64 However, viewing quality
children’s television programs, such as the
Sesame Street series, does improve cognitive
skills in pre-schoolers.65 The United States
spends more money per student on K-12 edu¬
cation than any other country. Nevertheless,
panels of business and government leaders
regularly decry the (alleged) fact that Amer¬
ican schools are failing to produce people
who can solve problems and face new cog-
62 Guerrero, 2009, Table 02.
65 Murray, 2003.
64 Zimmerman, Christakis, & Meltzoff, 2007.
65 Anderson, 1998.
nitive challenges - the very definition of fluid
intelligence.
The problem is not to show that the social
environment influences the development of
intelligence, the problem is to find out how
it does so. This is not easy.
Social variables are easy to name, but hard
to define. “Good parenting” is something we
all applaud, but just exactly what is a good
parent? Socioeconomic status (SES) is real.
But how do you measure it? Income is often
used as a proxy. In 2008 the Chief Justice of
the United States, John Roberts, had a salary
of $217,000. Alex Rodriguez, third baseman
for the New York Yankees baseball team,
had a salary of $28,000,000. Who had higher
socioeconomic status, the chief justice or the
third baseman?
Measurement is not the only problem.
As has been pointed out earlier, we do not
have a theory of the social environment that
approaches the clarity of Mendel’s theory
of genetic inheritance. Nor do we have any
broad theoretical approach to environmen¬
tal issues that can play a role similar to the
role that Darwin’s theory of evolution has
in biology. Lack of a comprehensive theory
of the environment has resulted in a great
many ad hoc studies without a great deal
of accumulated knowledge. The lack of an
adequate theory has also made it hard to
understand the relations that we do observe.
Social variables are highly collinear with
each other, with genetic measurements, and
with measurements of the physical environ¬
ment. Nutrition is linked to socioeconomic
status, especially in the developing world.
Parents who produce favorable home envi¬
ronments for their children are likely to seek
out the best school environments. In modern
developed countries (and in urban districts
worldwide) residence is closely tied to SES
and sometimes to ethnic status. This con¬
strains the composition of children’s play
groups. Without a theory it is difficult to
develop models that can guide the selection
of variables to study.
These are real problems. Nevertheless,
psychologists and educators have learned
something about social environments that
nurture or restrict intelligence.
284
HUMAN INTELLIGENCE
9.4.1. Socioeconomic Status and
Intelligence
Socioeconomic status (SES) refers to the
obvious, but nebulous, concept that there
are classes. There is no generally agreed-
upon definition of what a class is, so the
idea is clearly a fuzzy one. The commonest
practice is to define anywhere from three
to six different classes, based upon com¬
binations of income, education, and occu¬
pational prestige. It is not unusual to find
studies in which a single variable, such as
income or education, is used as a proxy for
SES. Given the variety of definitions, it is
a bit surprising to find that SES is corre¬
lated in the .30-.40 range with performance
on such diverse variables as the WAIS tests,
the SAT, and Raven’s progressive matrix
tests.66 However, the direction of causality
is far from clear, and it is probably bidi¬
rectional. There is a correlation of slightly
over .40 between parental SES and SAT
scores.6^ The same thing is true for other
measures of intelligence, such as IQ tests
and academic performance.68 These findings
suggest that SES is causal to intelligence.
At the same time, test scores obtained in
adolescence correlate with a person’s own
SES roughly twenty years later, which is
consistent with the argument that intelli¬
gence causes SES.69 As a further compli¬
cation, test scores and SES are both cor¬
related with parental advantages of various
sorts, including parental, and hence one's
own, genetic constitution. Parental advan¬
tage, carried forward across the generations,
could, at least in theory, be a cause for
one’s own socioeconomic success. And just
to make things even more confusing, cog¬
nitive tests such as the SAT are used as
screening devices in education, raising the
possibility that measures of intelligence act
as a gatekeeper for access to resources that
determine social success, but that intelli¬
gence itself has little causal influence.
66 Ceci & Williams, 1997; Raven, 1989.
67 Sackett et al., 2009.
68 Teasdale & Owen, 1986; Zwick & Green, 2007.
69 Herrnstein & Murray, 1994.
The problem is that SES is too global
a measure of either parental influence or
one's own success in life. In order to under¬
stand the influence of social class upon intel¬
ligence we have to take a finer look at
phenomena that underlie the correlation.
Three classes of studies have been used to
evaluate the effects of the social environ¬
ment: adoption studies, studies involving
social interventions, and multivariate anal¬
yses of specific features in the environment,
We consider each of them in turn.
9.4.2. Adoption
As we saw in Chapter 8, adoption studies
are frequently cited as supporting a genetic
cause for intelligence, on the grounds that
correlations between measures of cogni¬
tion in adoptees and their biological parents
are higher, and usually substantially higher,
than the correlations between adoptees and
their adoptive parents. A different statis¬
tic, the mean intelligence test scores of
adoptees, can be cited in support of envi¬
ronmental influences.
An early, much-cited British study in the
1930s produced what turned out to be fairly
typical results.70 The biological parents were
all “working-class," as defined in Britain at
that time. The mothers, as a group, had an
estimated mean IQ of 86. The adopting par¬
ents were all described as well educated.
Given this information, one would expect
the adoptees to have IQ scores in the 90s,
somewhat higher than the mothers’ IQs,
but still below the population mean of 100. 71
This is not what happened. The mean IQs
were 117 at age two, and 108 at thirteen. This
is evidence for an effect of the childhood
home environment. However, the correla¬
tions between the biological mother’s level
of education and adoptee's IQ scores were
70 Skodak & Skeels, 1949.
71 The reason for this is a statistical phenomenon
known as “regression toward the mean.” Whenever
an extreme score is observed on a test of less-than-
perfect reliability, the best estimate of the score that
would be obtained upon retesting, under exactly the
same conditions, would be a score between the orig¬
inal extreme score and the population mean.
ENVIRONMENTAL EFFECTS ON INTELLIGENCE
285
Table 9.2. Mean IQ scores of adopted children as
pre-schoolers and as adolescents, compared to scores
by biological children of the adopting family
Group Time 1 19 j6) Time 2 (^1986)
Biological children 116.4 109.4
Adopted African American 106.1 98.1
Adopted White 117.6 105.6
Source: Data excerpted from Weinberg, Scarr, & Waldman, 1992,
Table 2.
.04 at age two and .31 at age thirteen, reflect¬
ing what was to be a typical finding in later
studies: measures of genetic influences upon
intelligence rise as children grow older.
At this point, it may be helpful to look
back to Figure 9.1. If we substitute “adoptive
home” and “biological home” for “encourag¬
ing environment” and “restricting environ¬
ment,” the figure illustrates how an environ¬
mental effect upon mean scores could occur
along with a correlation between test scores
and measures of genetic potential.
We then “fast forward” almost forty years,
to the Minnesota Trans-Racial Adoption
(MTRA) study conducted in the 1970s in the
American Midwest. The MTRA study con¬
trasted the test scores of African American
and White children who had been adopted
into upper-middle-class White homes. Both
adoptees and biological children of the
adoptive family were tested, first when
the adoptees were pre-schoolers and then
ten years later, when the adoptees were
adolescents.72 Table 9.2 shows the results.
Biological children consistently outscored
adopted children in the transracial but not
in the intraracial groups. All test scores
declined somewhat from childhood to ado¬
lescence. This could be caused by a vari¬
ety of factors, including the fact that the
tests used differed somewhat over time. The
African American adoptees had an aver¬
age score near or above 100 (the putative
national mean) and well above the score of
85 typically found in African American pop¬
ulations on this type of test.
72 Scarr & Weinberg, 1976; Weinberg, Scarr, &
Waldman, 1992.
The practice of comparing the adoptees’
obtained scores to maternal scores or to
population expectations can be criticized.
Suppose that the birth mother’s IQ is 90.
This does not mean that a child would
be expected to have an IQ of 90, for
two reasons. No allowance has been made
for father's intelligence, which is typically
unknown. In addition, regression to the
mean (see footnote 71) implies that the
child’s IQ will be closer to the appropriate
population mean than is the mid-parent IQ.
But what is the appropriate mean? Moth¬
ers who give up their children for adoption
are hardly a randomly selected group of all
women, or even of all women in the appro¬
priate racial, ethnic, or educational group.
A better way is needed to estimate the
expected IQ of adopted children.
One way of doing this is to compare the
cognitive performance of adoptees to the
cognitive performance of children who have
not been adopted, but who might have been.
The test scores of adopted children can be
compared either to those of unadopted sib¬
lings or to unrelated, unadopted children in
the same pool of potential adoptees. When
this is done the results show a positive adop¬
tion effect of slightly less than one stan¬
dard deviation unit, both for test results
and for measures of school performance.73
This is probably an overestimate of the
adoption effect, because the comparison
includes the effects of any tendency lead¬
ing to adoption of apparently more favorable
children. However, given that most children
are adopted as infants, and that standard
75 Van Ijzendoorn, Juffer, & Poelhuis, 2005.
286
HUMAN INTELLIGENCE
Table 9.3. Mean WISC-R scores as a function of SES of birth and
adopting parents. The number in parentheses is the number of cases.
Adopting Parents Adopting Parents
High SES Low SES Average
Birth parents high SES 119.6 (10) 107.5 (8] 114.2
Birth parents low SES 103.6 (10] 92.4 (10) 98.0
Average 111.6 99.1
Source: Based on data reported by Capron & Duyme, 1989.
developmental inventories do not do a good
job of predicting later intelligence, this effect
is likely to be small. Except for pathological
cases, adoption agencies would have a diffi¬
cult time identifying infants who were going
to be bright or dull fifteen years later.
Understandably, most studies stress the
positive effects of adoption. Negative effects
are conceivable. It should be possible to
move intelligence upward or downward by
adoption, depending upon the relative SES
of the biological and adopting parents. Such
an analysis requires an unusual situation, for
high SES parents are less likely to give up
their children than are low SES parents, and
adoption agencies are more likely to place
adoptees with high rather than low SES
families. However, the unusual does occur.
French researchers were able to locate a rela¬
tively small number of young children (eight
to ten per cell) in a study that did fit the
appropriate design.74 Table 9.3 shows the
IQ scores achieved on the Wechsler Intel¬
ligence Scale for Children (Revised) at an
average age of fourteen years.
Taken at face value, the data in Table 9.3
indicates a high birth SES-low birth SES
effect of about 16 IQ points, and a high
adopting SES-low adopting SES effect of
12.5 points. In other words, the data is con¬
sistent with both hereditarian and environ¬
mental effects on intelligence, and makes
the important point that the two causes are
not mutually exclusive. I think it would be
unwise to go much beyond this, for the size
of the effects certainly should not be gener¬
alized. The study is a small one, and it is not
74 Capron & Duyme, 1989, 1996.
clear that the difference between the adop¬
tive high and low SES groups was equivalent
to the difference between the high and low
SES birth parents.
The results of adoption studies have been
used to argue for both environmental and
genetic influences on intelligence. People
who want to emphasize genetic causes cite
the fact that indices of adoptees' cognitive
competence are better predicted by bio¬
logical parent's competence than by adop¬
tive parent's competence as evidence for the
importance of genetics. People who want to
emphasize environmental causes cite gains
in intelligence achieved by adoptees. The
debate over how to interpret these find¬
ings can be heated. In some of the studies
put forward to support the genetic posi¬
tion the authors do not report changes in
mean scores, while in studies put forward
to support the environmental positions the
authors do not report parent-adoptee cor¬
relations. This practice more resembles the
behavior of a lawyer presenting the evidence
for a client than the behavior of a scientist
reporting data to be considered in evaluating
theories.
In fact, there is no conflict between the
results. Erik Turkheimer, of the Univer¬
sity of Virginia, offered an analysis that
brings both these results into the same
framework.75 His analysis was based upon
the concept of reaction range, as discussed
in Chapter 1, section 5, and illustrated in
Figure 9.1 of this chapter. Turkheimer devel¬
oped a mathematical model that separates
the genetic and environmental effects in
circumstances such as those illustrated in the
75 Turkheimer, 1991.
ENVIRONMENTAL EFFECTS ON INTELLIGENCE
287
figure, and used it to reanalyze a study con¬
trasting adopted and nonadopted siblings.76
The original authors had concluded that
there was a major effect of social class, on
the grounds that adoptees had higher intel¬
ligence scores than their nonadopted sib¬
lings, and that the adopting parents had
higher occupational status than the biologi¬
cal parents. According to Turkheimer’s anal¬
ysis, which treated the within-group and
between-group effects in a single frame¬
work, there was no reliable effect of a direct
measure of adoptive SES - father’s occupa¬
tional status - but there was still a large (and
unexplained) effect in favor of the adopted
children. How could this occur?
While various explanations of this finding
have been offered, I believe that the ambi¬
guities in studies like this will not be cleared
up until we take a finer look at environ¬
mental measures. Variables like “adoption,”
“socioeconomic status,” and even paternal
intelligence and occupation are distal vari¬
ables with respect to the development of
intelligence. Nutrition, schooling, and par¬
enting practices are proximal variables, act¬
ing directly upon intellectual development.
What we need to do is to look at some prox¬
imal effects.
9.4.3. The Home Entnronment
Parenting practices are evaluated by rating
homes on such things as the general order¬
liness of the home, the amount of read¬
ing or other explicitly educational material
available, the number and quality of interac¬
tions between parents and children, and the
extent to which children are encouraged to
work out the answers to questions and puz¬
zles, as opposed to being told how to do
so. It appears that the best environment for
intellectual development is one in which the
child is encouraged to work out problems
with guidance and support from parents, as
opposed to an authoritarian setting in which
the parent tells the child what to do, or
a laissez-faire setting in which the child is
pretty well left on his or her own. Parenting
76 Schiff & Lewontin, 1986.
styles, and especially substandard parent¬
ing practices, are statistically associated with
indices of socioeconomic status and family
solidarity, such as income and whether the
child is in a one-parent, father-absent, or
conventional mother-father home.77 These
variables are important, for the home envi¬
ronment has a great deal to do with intellec¬
tual development in young children. This
has been shown by two interesting lines of
research.
The first line involves observation of the
influence of the home environment upon
measures of children’s intelligence, includ¬
ing but not limited to test performance.
In one such study Victoria Molfese and
her colleagues at Southern Illinois Univer¬
sity followed 121 children, none of whom
had experienced extreme neonatal or birth
risks, from age three to age five.78 The
children were given intelligence tests annu¬
ally. Family SES was determined by com¬
bining indices of education, occupation,
and income. The home environment was
determined by observation and rating, using
the widely accepted Home Observation for
Measurement of the Environment (HOME)
scale, which requires observation of the
child’s home, including the provision and
use of reading material for children and
reports of the manner in which adults inter¬
act with the child.79 Figure 9.5 shows the
combined and independent contributions of
SES and HOME ratings to the prediction
of children's WISC scores. While there are
some irregularities, there is a trend toward
a decreasing influence of home environ¬
ment and increasing influence of SES as
the children aged. This is consistent with
behavior genetic studies that show generally
increasing influences of genetic heritage and
lowered influence of home environment
77 Kotchick & Forehand, 2002. Kotchick and Forehand
make the interesting point that if a family lives in a
potentially threatening environment, an authoritar¬
ian, controlling style of raising young children may
be adaptive, even though it does not foster intellec¬
tual development, because of the need to protect
the child.
Molfese, DiLalla, & Bunce, 1997. This article also
contains references to a number of related reports.
7? Bradley, 1993.
288
HUMAN INTELLIGENCE
— ♦— SES& HOME
SES:HOME
HOME:SES
Figure 9.5. The fraction of variance in WISC scores predictable
from measures of SES and home environment. For reference, a
fraction of .04 is equivalent to a correlation of .2. Calculations
based on data in Molfese, DiLalla, & Bunce, 1997, Table 2.
upon intelligence as people age, albeit over
a much greater time span than was the case
in the Molfese and colleagues study.
Two more studies illustrate the wide¬
spread influence of the home environment
during the early years. Both results come
from studies previously cited to make points
about the physical environment: observa¬
tions of rural Filipino children and observa¬
tion of children growing up in Kosovo, in the
Balkans. In the Philippine study, used earlier
to indicate the relation between nutritional
status and intelligence, ratings of home envi¬
ronments similar to the HOME scale added
to the prediction of a child’s intelligence,
after allowance for a variety of other vari¬
ables, including maternal intelligence test
score, paternal education, and physical envi¬
ronmental variables.80 In the Kosovo study
of atmospheric lead, the HOME score had
the highest correlation with children's intel¬
ligence test scores of all the variables consid¬
ered, which included measures of maternal
intelligence (Raven matrix score) and mater¬
nal education.81
The American Midwest, the rural Philip¬
pines, and the Balkans are very different
places. When relations between any variable
and intelligence are consistently related over
80 Church & Katigbak, 1991, Table 6.
81 Wasserman, 2000.
such diverse settings, one has to pay atten¬
tion. The microenvironment of the home
has a substantial influence on intelligence
during the early childhood years. A good
home environment gives a child a leg up
as he or she enters school. And, as will
be shown shortly, schooling counts as well.
Things mount up.
9.4.4. Early Childhood Interventions
in “at Risk” Populations
There have been substantial attempts to
improve the environments of children who
are thought to be “at risk” for showing
poor cognitive development. Since the 1960s
the US Government has funded Head Start
programs for pre-schoolers from low SES
families. Head Start was motivated by a con¬
cern that low SES, and particularly African
American, children entered primary school
lacking a number of cognitive and social
skills that were thought to be important in
children’s adjustment to the school experi¬
ence. These programs typically involve pre¬
school training for a half-day, five days a
week, for the year prior to entering school.
Many Head Start programs also include
some form of parental education, as it is felt
that inadequate parental support may be a
factor leading to the children's poor perfor¬
mance in schools.
ENVIRONMENTAL EFFECTS ON INTELLIGENCE
289
Progress toward the goal of improving
cognition has been spotty, at best. In 1969
Arthur Jensen, a Professor of Education at
the University of California, Berkeley, pub¬
lished a controversial paper in the Harvard
Educational Review. He began
Compensatory education has been tried
and apparently it has failed.
Jensen, 1969, p. 2
Later in the article he said
The evidence so far suggests the tentative
conclusion that the payoff of preschool and
compensatory programs in terms of IQ
gains is small.
And further down on the same page:
The techniques of raising intelligence per se
in the sense of g, probably lie more in the
province of the biological sciences than in
psychology and education.
Jensen, 1969, p. 108
Jensen maintained this position, vir¬
tually without change, in an important
book on intelligence published almost thirty
years after the Harvard Educational Review
article.82
A similar conclusion was echoed in 1994
by Herrnstein and Murray, in a book that
was received as contentiously as Jensen’s
earlier conclusions had been.83
The school is not a promising place to try
to raise intelligence or to reduce intellec¬
tual differences, given the constraints on
school budgets and the state of educational
science.
Herrnstein and Murray, 1995, p . 414
Quite a different view has been expressed
by Edward Zigler, a Yale Professor who,
as a government official, became famous as
the “Father of Head Start.” Zigler observed
that a variety of behaviors are improved by
Head Start programs. These include good
study habits, cooperative work skills, and
improved parental support. Improving these
characteristics was, according to Zigler, at
82 Jensen, 1998.
83 Herrnstein & Murray, 1994.
least as important as improving the cogni¬
tive skills measured by test scores.84
The debate over Head Start is more
nuanced than it appears when people sim¬
ply recite mantras that Head Start programs
do (or don’t] work. Critics like Jensen and
Herrnstein and Murray generally focus on
IQ and similar tests as essentially complete
measures of intelligence, while the support¬
ers have a more expanded view of intelli¬
gence, to include things like knowing how
to manage time and how to learn arbitrary
material - “associative learning," in Jensen’s
terms. The supporters also stress improve¬
ment in the child’s general social situation,
as Zigler's comment about parenting skills
indicates. Such concerns, which strike me
as being quite legitimate criteria by which
to evaluate a pre-school program, are not
considered in the criticisms raised by Jensen
and Herrnstein and Murray.
There is another qualification. Jensen and
Herrnstein and Murray did not say that pre¬
school programs will not work; they said
that those programs that were economically
realistic did not work. That is an important
distinction. “What early childhood programs
improve intelligence?” is a question for edu¬
cational psychologists. “What could work at
a cost of x dollars per child?” is an important
question for educational policy makers. The
two viewpoints are not the same. In his book
on the improvement of intelligence, Nisbett
concluded that
Several early childhood education pro¬
grams actually do produce large immediate
gains in IQ, as well as long term gains in
IQ or academic achievement, or both.
Nisbett, 2009, p. 120
Two meta-analytic reviews85 show that
the picture is not as bleak as Jensen painted
it, but that Nisbett’s statement may be a bit
optimistic.
There is a great deal of variation among
pre-school intervention programs. Expendi¬
tures cover an eight-to-one ratio. At the low
end we have minimal head-start programs,
84 Zigler & Styfco, 1997.
85 Barnett, 1998; Gorey, 2001.
290
HUMAN INTELLIGENCE
in which children attend reasonably well-
run pre-school programs with educational
components from two to five times a week
for one or two years. At the high end some
intense programs have included staff/client
ratios as high as one well-trained teacher
for every three students, parental counsel¬
ing, and interventions for five years or more.
In general, the meta-analyses show that
the less intense programs have short-term
results, but that these results (in terms of
IQ scores and school accomplishment) fade
away quickly. By contrast, the more intense
programs have results that last for years, in
terms of both intelligence test scores and,
more importantly, school achievement. One
meta-analysis reports an improvement of
nine IQ points for the intensive programs,
five years after the programs ended. This
would be somewhere toward the end of ele¬
mentary school for most participants. Only
a few studies have followed students as far
as high school, although two very intense
intervention studies traced participants into
adulthood. These studies report substantial
positive effects on social behaviors, such as
encounters with the law, and smaller effects
on IQ scores more than ten years after pre¬
school participation.
Possibly the most intensive program, the
ABCDerian project, was aimed at a deeply
impoverished group of students in North
Carolina. This study has reported positive
results, compared to a control group, at age
twenty- one. 86 Because this is generally con¬
ceded to be one of the most effective (and
expensive) of the pre-school programs, it is
worth looking at the study in more detail.
The participants were children from low
SES families in North Carolina. Virtually
all the children were African American.
Slightly over one hundred children were
assigned to the special program or to a
control group. The intervention began at
an average age of 4.4 months and contin¬
ued until the children entered kindergarten.
Depending upon the period, there was one
instructor for every three to six children.
In addition to instruction and supervision,
86 Campbell et al., 2001.
a nutritional program was developed and
offered to both the experimental and con¬
trol group. The pre-school lasted virtually
the entire day, five days a week. Participants
were followed until they were twenty-one.
When the program ended, at age five, the
children took the Wechsler Preschool Intel¬
ligence test. The experimental group had
a mean IQ of 100 and the control group
a mean IQ of 94. (Both scores are some¬
what above what would be predicted from
demographic data.) Figure 9.6 shows the
results of subsequent testing. There are two
striking features of the graph. Both groups
showed a steady decline in test scores over
the years after leaving the project. At the
same time, the experimental group main¬
tained a roughly five IQ points ( d = .3)
advantage over the control group through¬
out. How could this come about?
The decline is not surprising. IQ tests for
children below the age of five are at best
moderately predictive of scores at a later
age. This could be because the tests for
very young children are tests of cognitive
functions different from those evaluated by
later tests, or it could be because the cogni¬
tive systems important for test performance
(e.g., working memory) are only slightly
developed in young children. (Recall that
heritability coefficients are lowest for very
young children, suggesting that maturing of
the cognitive system is a serious possibility.)
Several studies, to be discussed in Chapter
11, have shown that African American chil¬
dren obtain test scores in the 100 range
(equivalent to Whites) in the early school
years. African American scores decline to
a mean of roughly 85 by adulthood. The
control group appears to be doing this. The
experimental group also showed a regression
effect, but maintained its advantage over the
control group into adulthood.
The picture for academic achievement,
which is of more social importantce than
doing well on a test, is more encouraging.
One meta- analysis87 determined that 80% of
the children who participated in intensive
programs had a higher level of achievement
87 Barnett, 1998.
ENVIRONMENTAL EFFECTS ON INTELLIGENCE
291
•- ABCDerian
— S- Control
Figure 9.6. Mean IQ scores obtained by the participants in the
ABCDerian project and in a randomly chosen control group. The
intervention ended when the children entered school, at
approximately age six. Data from Campbell et ah, 2001, Table 1.
than that of the median participant in the
control group, five years or more after the
intervention ended. Participants also experi¬
enced fewer social and behavioral problems.
Entry to college was also higher in partici¬
pants of these programs, compared to con¬
trol groups.
So what does the scoreboard say? State¬
ments like “Head Start is a (failure) (suc¬
cess)” mask agreement over the facts, and
disgreement over how to describe them. If
you believe that the role of early child¬
hood intervention and special education is
to improve intelligence, as measured by
test scores, the critics are right that at
best marginal improvements can be made.8”
However, the programs do improve a num¬
ber of academically relevant behaviors, rang¬
ing from study skills to improvement in
interacting with students and teachers. Such
behavior leads to improved learning in for¬
mal educational settings. These are not neg¬
ligible benefits.
Scaling up a project like the ABCDerian
project to cover a substantial portion of low
SES children would require a huge financial
investment and pose a tremendous problem
of staffing. One can argue over the relative
values of the costs and benefits, and how
they should be assessed. These are educa¬
tional policy issues, not issues in the study
of intelligence.
9.4.5. The Home Environment:
Competition for Resources
The behavior genetic studies reviewed in
Chapter 8 have consistently found that
‘nonshared environmental influences/ which
refers to environmental influences that dif¬
fer between family members, are a sub¬
stantial source of environmental effects on
intelligence/9 As many autobiographical
accounts make clear, one of the biggest
differences in a child's early life is sim¬
ply whether he or she is raised in a large
family. The extreme, I suppose, would be
a medieval royal family, where early-born
children, as potential heirs to the throne,
received very different treatment than did
later-born children. As anyone who has
either raised or been raised in a large fam¬
ily will attest, having several brothers and
sisters is not like being an only child.
Children raised in large families tend, on
the average, to have lower test scores than
88 Detterman & Thompson, 1997; Herrnstein &
Murray, 1994; Jensen, 1998.
89 Jensen, 1997.
292
HUMAN INTELLIGENCE
Panel 9.9. The National
Longitudinal Studies
The National Longitudinal Studies are
studies carried out by the US Depart¬
ment of Labor in order to obtain an accu¬
rate picture of the demographics, social
structure, and economics of the United
States. Such information is needed by
policy makers. The surveys also provide
valuable sources of data for economists,
sociologists, educational researchers, and
increasingly for psychologists. In general,
the surveys utilize a random sample of
US citizens relevant to the purpose of the
survey. On occasion certain subgroups of
interest are intentionally oversampled, in
order to provide more accurate informa¬
tion about them.
The National Longitudinal Study of
Youth 1979 (NLSY79) was a survey of
over 10,000 young men and women, aged
fourteen to twenty-two when the sur¬
vey was initiated in 1979. This sample
has been followed up periodically. Infor¬
mation on health, economics, and social
status has been obtained. At the time
of the first survey a fortuitous event
occurred. The Department of Defense
needed to update its normative sample
of the Armed Services Vocational Apti¬
tude Battery (ASVAB) and the associ¬
ated Armed Forces Qualification Test
(AFQT). Accordingly, a large percent¬
age of the NLSY79 participants took
the ASVAB. The result was an impor¬
tant prospective study of intelligence,
for the ASVAB scores of the fourteen-
to twenty-two-year-old participants can
be related to their subsequent progress
through life. Herrnstein and Murray
made extensive use of the NLSY79
database in their provocative 1994 book.
The NLSY has "spawned” (in a some¬
what literal sense] a second survey, the
NLSY Children and Young Adults study,
in which data is gathered on children
of the NLSY79 participants. This pro¬
vides social scientists with valuable data
on changes in the population over gener¬
ations.
The NLSY97 survey is something of
a repeat of the 1979 survey. Over 9,000
young men and women born in the 1980-
84 period were enrolled, and are inter¬
viewed on an annual basis. The pur¬
pose is to track social issues concerned
with the transition from youth to adult¬
hood. Comparisons to the NLSY79 data,
when possible, provide a way of compar¬
ing cohorts as they transit from youth to
adult life.
children raised in small families. This trend
is shown in Figure 9.7, which presents data
from the National Longitudinal Study of
Youth 1979 study (described in panel 9.9].
Similar findings have been obtained in many
other studies. There is a correlation some¬
where between —.15 and —.20 between
family size and children’s intelligence test
scores.90
While the fact is clear, the reason for
the fact is definitely not clear. Most debates
stress either genetic or environmental influ¬
ences. Galton worried that because large
families seemed to produce less intelligent
90 Anastasi, 1956; Herrnstein & Murray, 1994; Lynn,
1998; Lynn & Harvey, 2008; Lynn & Van Court, 2004.
people the intelligence of the society, as a
whole, must fall. Essentially the same con¬
cern over a "dysgenic effect” was voiced
by Cattell in the mid twentieth century,91
and by Richard Lynn more than fifty years
later.92 A key point in their argument is
that family size increases as maternal intel¬
ligence drops, as is shown in Figure 9.8. The
data shown are typical of numerous similar
findings.
Those who worry about dysgenics, from
Galton to Lynn, appear to assume that
graphs such as these represent a proximal
genetic effect; unintelligent mothers have
91 Cattell, 1940.
92 Lynn, 1998.
ENVIRONMENTAL EFFECTS ON INTELLIGENCE
293
Figure 9.7. Mean intelligence estimates of young children born to
mothers who were in the NLSY79 survey. The estimate is based on
the Peabody Test, which is suitable for children in the pre-school
and elementary school age range. Based on Rodgers et ah, 2001,
Table 4.
children who are genetically unintelligent,
and have lots of them. Before we accept
this argument a third phenomenon has to
be considered, the birth-order effect.
Intelligence test scores are related to
birth order; firstborns, on the average, have
higher scores than second-borns, and so on.
Figure 9.9 shows the results of a study
from Norway, in which birth-order effects
were shown in the intelligence test scores
obtained by Norwegian young men when
they registered for military service.93 The
birth-order effect is clearly social. If an older
sibling dies, the next sibling in line assumes
the “benefits” of the older sibling’s place.
Evidently acquiring intelligence is a bit like
the medieval rules for acquiring a kingship:
oldest surviving child gets the crown and the
IQ points.
Birth-order effects are not apparent in the
NLSY data set (Figure 9.7). This is not a
contradiction of the evidence for birth-order
effects, for the children of NLSY partici¬
pants were tested before they were ten. If
93 Bjerkedal et al., 2007; Kristensen & Bjerkedal, 2007.
birth-order effects are based on a competi¬
tion for resources, the effect may accumu¬
late over time, rather than being apparent
early in the child’s life. At the same time, the
birth-order effect cannot be the sole cause
of the family-size effect, for, as Figure 9.7
shows, the family-size effect can be obtained
in a situation where there is no birth-order
effect.
Robert Zajonc, a professor at the Uni¬
versity of Michigan, and his colleagues have
developed a model of family dynamics that
they refer to as the confluence model.9*
Zajonc argues that a child’s intelligence will
be stimulated by an intellectually challeng¬
ing environment, and that the mean age of
the family is an indicator of the intellec¬
tual challenge in the home environment. To
illustrate, a seven-year-old only child whose
father and mother are both thirty will be
in a family where the mean age of family
members is (30 + 30 + 73/3, or 22 1/3. For a
seven-year-old with a three-year-old sibling,
the mean age of family members is (30 +
30 + 7 + 33/4, or 17 V2. This accounts for the
94 Zajonc, 1983; Zajonc, Markus, & Markus, 1979.
HUMAN INTELLIGENCE
294
Maternal AFQT
Figure 9.8. The relation between number of children and maternal
intelligence score (AFQT percentile score) in the NLSY79 data set.
Source: Rodgers et al., 2001, Table 6.
family effect; large families will have lower
mean ages of the members.
The general tone of Zaj one’s argument,
that large families do not provide good envi¬
ronments for the development of intelli¬
gence, has been widely accepted. It has even
led to 'popular psychology” advice against
having large families, on the grounds that
the social environment of a large family will
work against the development of children’s
intelligence.95
The facts are clear; the explanation is not.
Here are three alternatives.
1. A proximal dysgenic hypothesis: Moth¬
ers’ test scores indicate their genetic
potential (probably in part correct), and
they pass on this potential to their off¬
spring (certainly true). The offsprings’
test scores reflect the offsprings’ genetic
potential (probably in part correct).
Genetic potential acts as a proximal
influence on intelligence, thus produc¬
ing the family-size effect.
2. A distal genetic hypothesis: Low mater¬
nal intelligence results in poorer parent¬
ing practices, including a tendency to
begin child bearing at an earlier age.
This practice may itself be partially
due to genetic influences. The environ¬
ment in large families tends to restrict
the development of intelligence. Genet-
95 See, e.g., Brothers, 1981.
ics acts as a distal variable, leading to
environmental practices that, as prox¬
imal variables, directly influence the
intelligence of offspring.
3. An environmental explanation: The
following argument is based on Nis-
bett’s stress that culture matters in the
development of intelligence.96 Mater¬
nal intelligence test scores are negatively
correlated with SES, and are also cor¬
related with membership in low-status
racial and ethnic groups. These groups
have reduced access to resources in time
and money that could improve parent¬
ing practices. In addition, there is a
tendency for them to follow cultural
practices that lead to childbearing at a
younger age, and therefore to produce
large families in which children have to
compete for resources.
An argument based on proximal genetic
influences, alone, cannot be maintained
because it cannot account for the birth-order
effect. It also cannot account for the world¬
wide trend toward reduced family sizes as
people move into urban settings.97
Either the distal genetic or environmen¬
tal explanations could account for all three
effects: family-size, maternal intelligence,
and birth order. As is often the case with
social and behavioral phenomena, multiple
96 Nisbett, 2009.
97 Mace, 2008.
ENVIRONMENTAL EFFECTS ON INTELLIGENCE
295
107
106
_ 105
O
£ 104
LO
CT)
g 103
§ 102
g
101
100
99
98
Figure 9.9. The birth order effect appears in families where a
child becomes the oldest due to the death of an elder sibling.
Scores have been adjusted to allow for parental education
level, maternal age at birth, birth weight, and cohort effects.
Bars show 95% confidence intervals. From Kristensen, P., &
Bjerkedal, T. (2007) Explaining the relation between birth
order and intelligence. Science, 316 (5832], 1717, Figure 1.
Reprinted with permission from AAAS.
♦ No siblings died
♦ One sibling died
♦ Two siblings died
m •
♦ 4
►
I
3E
►
1 2 3
Birth order
explanations are possible, and they are often
not mutually exclusive.
Family-size effects, along with the
cohort effect and the repeated findings
of large heritability coefficients, represent
well-documented phenomena with unclear
explanations. Psychological research has
established what happens, but not why it
happens.
9.4.6. Summary and a Value Judgment
Family environments clearly do exert an
effect on children’s development of intel¬
ligence. Within the range of normal fam¬
ily environments in the developed nations,
roughly within the top two-thirds of
the socioeconomic spectrum, these effects
appear to be modest and ephemeral. They
virtually disappear in adulthood. The sit¬
uation is quite different when we look at
extreme environments, such as the homes
of the children at risk for lowered cognitive
development - for example, those who par¬
ticipated in the ABCDerian project.
Unfortunately, though, this does not get
us very far, for SES is a composite, abstract
variable that covaries with many other
variables. It is difficult to disentangle the
effects of family social practices - such as
failure to encourage children's problem solv¬
ing, authoritarian styles of adult-child inter¬
action, and other practices considered bad
parenting - from concomitant deficiencies in
the physical environment and genetic inher¬
itance. We cannot unambiguously assign
damaging SES effects either to heredity, or
to social or physical variables. Just as there
may be many genes affecting intelligence,
each one individually making a small contri¬
bution, there may be many familial variables
affecting intelligence, each one individually
making a small contribution.
High-cost intervention projects have pro¬
duced improved cognitive performance.
Unfortunately, the effects are far less than
most social engineers would like to see.
Many less expensive projects, such as the
typical Head Start program, show very lit¬
tle effect on test scores, or on other purely
296
HUMAN INTELLIGENCE
cognitive behavior, beyond the early school
years. However, there are indications that
these programs do prepare at risk children
for the social experience of school, and that
this may facilitate later school development.
The last point is very important. In indus¬
trial and post-industrial societies schooling
is a major factor in determining a person’s
contribution to the general society. In 2006
it was estimated that in the United States
a person without a high school diploma
earned only two-thirds as much a someone
with one.98 Other social problems, such as
poor health and criminal convictions, also
are negatively correlated with the possession
of a diploma. Any program that increases
the likelihood that children in at risk situ¬
ations will complete schooling is a socially
valuable program. It has also been estimated
that even the more expensive programs are
cost-effective, when the total cost to society
over a long period of time is considered.99 I
must admit, though, that these projections
are based upon a number of assumptions
that I, for one, find suspect.
I close this section with a comment that is
avowedly, and openly, a statement of social
beliefs.
I once heard an address by Sandra Scarr,
herself a major contributor to the litera¬
ture on genetic and environmental effects
on intelligence, and by no means a person
who denies the importance of genetics. I
regret that I have no reference except my
memory. Scarr pointed out that a child born
to a poor family, with limited resources,
is often in a bleak environment. It is a
good thing, in itself, to provide programs
offering such children better nutrition, a
safe place to interact with adults and other
children, and an interesting, challenging
environment. If the resulting program also
improves their cognitive abilities, their intel¬
ligence in the important, conceptual sense,
then that is a very nice benefit of having done
a good thing. If the program just improves
test scores, but does not improve cognitive
98 Retrieved from centerforpubliceducation.org, June
2008.
99 Barnett, 1998.
abilities in the more general sense, then that
is just a curiosity. Whatever the outcome,
society should provide as good a social envi¬
ronment as can be arranged. You do not
need a reason to do a good thing, you should
just do it.
I find Scarr's reasoning compelling. Pro¬
viding aid to disadvantaged children is a duty
owed to the children. Costs and benefits are
relevant when comparing programs; there is
no sense in spending more than you have to.
However, I believe that society has a duty
to provide such programs, just as much as it
has a duty to provide military defense and
security for the aged.
9.5. Education
The cohort effect shows that changes in the
environment can result in major changes
in intelligence, on a population basis. Over
the twentieth century the industrially devel¬
oped countries saw improvements in nutri¬
tion, better health practices, smaller family
sizes, and increases in the availability of pre¬
school programs. There have also been huge
changes in education.
The US data, which mirrors that of other
industrially developed countries, shows how
strong the educational change has been. In
1900 the school enrollment rate for fifteen-
to nineteen-year-olds was 50%; it rose to 75%
by 1940, and has remained relatively stable
at slightly above 90% since 1990. In 1910 the
median number of years of education for
people twenty-five years old or older was
8.1 years. In 1940, at the outset of World
War II, it was only 8.6. In 2008 better than
85% of adults had twelve or more years of
education.100 In the first half of the twentieth
century the big change in education was that
more people gained basic skills in the tra¬
ditional “reading, writing, and ‘rithmetic”’
than had been the case earlier. Illiteracy,
arguably the most important single result of
a lack of formal schooling, had fallen from
10.7% in 1900 to 2.9% in 1940. By 2000 the
100 U.S. Census Bureau, 2010 Statistical Abstract,
Table 224.
ENVIRONMENTAL EFFECTS ON INTELLIGENCE
illiteracy rate was less than 1%. The second
half of the twentieth century was character¬
ized by marked increases in secondary and
tertiary education, especially for women and
minority group members. By 2007, 29% of
the adults in the US had a college degree.
Similar trends have been observed in other
industrialized countries. It is not unreason¬
able to assume that all this education had
an effect on cognitive skills, and hence, on
intelligence. But just what?
Once again we have to be clear about
our definitions. If we take the attitude that
intelligence is essentially g, as indexed by
a g-loaded nonverbal test, such as a pro¬
gressive matrix test, education may not
increase intelligence that much. Alterna¬
tively, if we accept the fluid intelligence-
crystallized intelligence (Gf-Gc) distinction,
we may find different effects of education
on Gf and Gc. I will generally take this
approach, as I believe that it offers a use¬
ful perspective for dealing with the effects
of schooling. I will keep coming back to
my point that the important components of
intelligence are those skills that are useful in
society. Test scores are of interest only to the
extent that they indicate which examinees
possess these skills.
We also have to be careful about what we
mean when we talk about education. Do we
mean any form of training? Do we include
commercial programs that are intended to
increase one’s intelligence (many are adver¬
tised, few have been evaluated), or do we
restrict our interest to programs of formal
education, the familiar K-12 and college/
university systems? I will consider two issues:
the influence of the formal system and the
(very few) attempts that have been made
to teach general reasoning skills and that
have been formally evaluated. And where
do commercial programs to increase scores
on tests like the SAT fall on this spectrum?
9.5.1. What the Educational System
Tries to Do
The developed countries hand over a great
deal of the task of educating the young to
formal school systems. This is a marked con¬
297
trast to the situation less than 300 years
ago, when most children learned what they
needed to know by some form of appren¬
ticeship, augmented by a very small amount
of formal training. Formal schools did exist
in both pre-literate and ancient societies.
The seafaring Polynesians had schools to
teach navigators to guide canoes on ocean
voyages of up to a thousand miles. As
part of their training they learned to use
sophisticated star charts.101 In ancient Egypt
scribes went to school to learn account¬
ing, surveying, construction, mathematics,
and engineering.102 What is different about
our society is the near-universality of formal
schooling, not the concept itself.
Today’s K-12 school systems are responsi¬
ble for transmitting three classes of knowl¬
edge: the basic beliefs and traditions of the
society, cognitive and motor skills useful
in solving frequently encountered problems,
and specific pieces of knowledge that will be
required in utilizing these skills.103 To illus¬
trate, somewhat flippantly, for the United
States circa 2010, this means, “God Bless
America, learn to read, and a camel’s a
mammal.”
Upwards of 50% of the citizens in
the industrially developed nations receive
some form of post-secondary education.
While there are some discussions of
the importance of a liberal education,
most post-secondary instruction is oriented
toward the training of specialists in fields
ranging from welding to the law. The result
has been a tremendous extension of the time
between the end of childhood and the point
at which a fully trained adult enters soci¬
ety. Alexander the Great was twenty-two
when he led the Macedonian/Greek army
into Asia. Napoleon was appointed brigadier
general at twenty-four. In today’s military
they would both be lieutenants.
Richard Snow, an educational psychol¬
ogist at Stanford University, has argued
that schools are successful if they provide
101 Hutchins, 1983.
102 Information provided by the King Tutankhamen
exhibit at the Field Museum, Chicago, 2006.
103 Cole, 2005.
298
HUMAN INTELLIGENCE
students with useful aptitudes, by which he
meant skills that can be used to operate
in the larger society and to guide further
learning. 1Q4 Snow divided the cognitive apti¬
tudes into two classes: specific skills (e.g.,
reading, knowledge of history) used to solve
certain classes of problems, and general rea¬
soning skills used both to solve problems and
to guide further learning. Snow's categoriza¬
tion of aptitudes mirrors the Cattell-Horn
distinction between Gf and Gc, but was set
in a broader context than that of the testing
paradigm.
Snow pointed out that schools also taught
skills in self-management and cooperative
problem solving. While some may not wish
to include such skills in the definition of
intelligence, they are certainly relevant to
the application of one's intelligence to prob¬
lems outside of a school setting. These
skills are often taught implicitly, by proce¬
dures that Snow referred to as a metacur¬
riculum, encouraging independent inquiry
and cooperative studies. This is certainly
the approved practice in education today,
although it is not always what actually hap¬
pens. It represents a departure from the
authoritarian, didactic methods that were
common in all schools up to about 1950, and
that are not unknown today.105
9.5.2. The Evidence for Educational
Influences on Intelligence
The American President Theodore
(“Teddy”) Roosevelt is reputed to have
said,
A man who has never gone to school may
steal from a freight car; but if he has a uni¬
versity education, he may steal the whole
railroad.
Attributed to Roosevelt by Laurence
Peters (1977, p . 117)
Can you imagine a more ringing affirmation
of the value of education?
I suspect that Roosevelt would have
regarded the question “Does schooling
104 Snow, 1996.
105 Bransford, Brown, & Cocking, 1999.
increase intelligence” as trivial, because it is
obvious that educated people are generally
capable of solving problems that uneducated
people cannot. Many of these problems are
socially relevant. They range from reading
and understanding newspapers to balancing
checkbooks and comprehending the terms
of mortgages. A substantial part of a per¬
son's intelligence, in the conceptual sense of
being able to solve socially relevant prob¬
lems, is clearly the product of education.
What about the effects of education upon
intelligence in the much narrower sense of
improving test scores? Today that is a fairly
hard question to answer, because virtually
everyone goes to school at least up to the
point at which education shifts toward spe¬
cialty training. However, this was not always
true, so prior to World War II it was possi¬
ble to conduct a study contrasting groups
of children, of similar socioeconomic back¬
ground, who either did or did not have
access to formal schooling. This situation
often occurred for reasons that had nothing
to do with the personal traits of the students,
such as the decision to locate a road, and
with it a school, near one community and
distant from another. The children who had
access to schools had higher intelligence test
scores than those who did not. The differ¬
ence in test scores between groups increased
with increases in the difference in the avail¬
able schooling.106
In today's society there is a positive rela¬
tion between a person’s intelligence test
score and his or her level of education. High
school drop-outs have generally lower intel¬
ligence test scores than those who complete
high school. The effect interacts with SES;
low intelligence is more predictive of drop¬
ping out for students from low SES fami¬
lies than it is for students from moderate or
high SES families.107 But what is cause and
what is effect? The correlation could come
about because intelligence produces success
in school, or because increasing amounts of
schooling produce intelligence, or both.
106 Ceci, 1990; Bronfenbrenner et al., 1996; Ceci &
Williams, 1997.
107 Herrnstein & Murray, 1994, Chapter 6.
ENVIRONMENTAL EFFECTS ON INTELLIGENCE
299
Figure 9.10. The path diagram for a study
showing the influence of education on
intelligence. It is necessary to show that
education has an influence on intelligence at
time 2 (the ? in the figure) in addition to the
influence exerted by intelligence at time 1.
One way to disentangle the situation is
to look at cases in which nonschool intelli¬
gence test measures are taken before some
variable period of schooling, followed by a
second intelligence test. The path diagram
for this design is shown in Figure 9.10. The
question is whether scores on the second
intelligence test are influenced by schooling,
after accounting for the relation between the
first and second tests. Herrnstein and Murray
analyzed scores for people in the NSLY79
for whom first test scores were available
(see panel 9.9) and concluded that schooling
had little, if any, effect. This result has been
widely quoted. However, a detailed reanaly¬
sis of the same data by Christopher Winship
(Harvard University) and Sandor Korenman
(City University of New York)108 questioned
their conclusion. Winship and Korenman
also reviewed a number of other studies
using some variant of the design sketched
in Figure 9.10. A Norwegian study was par¬
ticularly important, because it provided an
unusually good natural experiment.
Norwegian schoolchildren are given intel¬
ligence tests at age thirteen. At age eighteen
all Norwegian men take an intelligence test
as part of their registration for military ser¬
vice. The eighteen-year-old population con¬
tains some people who are still students,
and others who have dropped out at various
periods following completion of compulsory
108 Winship & Korenman, 1997.
education. Those who continued as students
tended to have higher test scores at age eigh¬
teen than did the registrants who dropped
out of the educational system. This fact,
alone, is not compelling evidence, because it
could be because the more intelligent people
stayed in school longer. However, the peo¬
ple who stayed in school had higher scores
at age eighteen than would have been pre¬
dicted from their scores at age thirteen, indi¬
cating a beneficial effect of schooling.
On the basis of their own analyses and
the studies that they reviewed, Winship
and Korenman estimated that an additional
year of education (through the K-12 years)
adds approximately 2.7 IQ points to a per¬
son’s test score. Speculating a bit, suppose
we combine this estimate with data on the
increase in educational attainment in the
United States. In 1946 the median of edu¬
cational attainment for adults was 8.6 years.
Today the median is “some college,” which
I arbitrarily (and conservatively) set at
13.1 years. Most of this increase has occurred
since 1940. Taking the Winship and Koren¬
man figures at face value, this implies that
a 4.5 x 2.7 = 12.15 IQ P°int rise in test
scores over the last half of the twentieth cen¬
tury could be accounted for by increases in
education.109
What does education do to raise intelli¬
gence? We need to distinguish between apti¬
tudes that are taught explicitly and those
that are taught implicitly.
Explicitly, educational systems try to pro¬
vide students with the knowledge and cog¬
nitive skills required to operate in their soci¬
ety. This is very close to the definition of
Gc. Not surprisingly, all battery-type intel¬
ligence tests include tests that assess such
knowledge and skills, either directly or indi¬
rectly. This is certainly true of the Armed
Forces Qualifying Test (AFQT) described
in Chapter 3 of this book, and treated by
some researchers (including Herrnstein and
Murray) as virtually synonymous with g.
109 Nancy Robinson, a developmental psychologist at
the University of Washington, has pointed out to
me that this argument does not explain why scores
on young children's intelligence tests have also risen.
300
HUMAN INTELLIGENCE
The AFQT is based on the Armed Services
Vocational Aptitude Battery (ASVAB] sub¬
tests of word knowledge, arithmetic reason¬
ing, paragraph comprehension, and math¬
ematics knowledge - all subjects that are
explicitly taught in school. As a result, the
ASVAB, and of necessity the AFQT, evalu¬
ates Gc.uo
Schools implicitly teach three other apti¬
tudes that are important in modern society:
problem solving removed from the imme¬
diate context of life, abstract conceptual
thinking, and (more recently) the ability to
detect patterns in stimuli. There is no claim
that the teaching is perfect, or that schooling
is the only way to develop reasoning skills.
There are many illustrations showing that
relatively unschooled people can exhibit
sophisticated reasoning in the context of
everyday problems.111 These cautions do not
detract from the main point. All that needs
be maintained is that schooling increases the
likelihood that children will acquire power¬
ful, abstract problem-solving methods that
can be applied to a variety of problems.
Literacy is probably the most power¬
ful of these skills. Studies contrasting lit¬
erate and nonliterate societies have shown
that the possession of literacy increases will¬
ingness to think about hypothetical situa¬
tions, and that literacy increases the ten¬
dency to use abstract classification systems
based on object features, rather than clas¬
sification based upon a concrete situation.
Classification-based reasoning tasks are part
of many intelligence tests. In addition,
modern educational methods have stressed
the importance of evaluating evidence by
detecting patterns. This is particularly the
case for instruction in science and mathe¬
matics. Spearman argued that this ability -
“eduction,” in his terms - is one of the most
important components of intelligence.112 An
analysis of the techniques used in modern
elementary school mathematics classes has
shown not only that the schools teach "educ-
110 Roberts et al., 2000.
111 See Cole, 2005; Hunt & Minstrell, 1994, 1996; and
Lave, 1988, for some examples.
112 Spearman, 1923.
Figure 9.11. A problem similar to the problems
presented in K-2 mathematics textbooks. The
task is to complete the sequence of figures.
Based on examples provided in Blair et al., 2005.
tion,” but also that some of the exercises that
they use to do so closely resemble the sorts
of items that appear on intelligence tests of
inductive ability. ll5 An example is shown in
Figure 9.11.
The relation between intelligence and
schooling is interactive. Not everyone will
gain exactly 2.5 IQ points from a year of
education. There are strong feedback loops
between intelligence and the effects of edu¬
cation; more intelligent people learn more,
and often quite a bit more, from schooling
than do less intelligent people. We go into
this in some detail in Chapter 10. Neverthe¬
less, the overall message is clear. Education
does increase intelligence, both in the nar¬
row sense of increasing IQ scores and in the
far more important sense of increasing one's
ability to solve life’s problems.
Was Roosevelt’s praise of university edu¬
cation justified? Although some university
officials claim that the purpose of a (lib¬
eral) college education is to teach people
to think, the claim has never been tested.
This is due, in no small part, to university
educators’ failing to agree on just what the
empirical definition of “being able to think”
ought to be. Of course, we do know that
colleges and universities do an excellent job
of training high-level specialists such as engi¬
neers and physicians. And for a direct test of
Roosevelt’s claim?
In the first six months of 2009 two sepa¬
rate financial scandals were revealed. Quite
independently of each other, Bernard Mad-
off and Allen Stanford, both university grad¬
uates, had bilked investors of tens of billions
of dollars in sophisticated frauds. This makes
Robin Hood look like a penny ante thief, and
makes Roosevelt look prescient.
Blair et al., 2005.
ENVIRONMENTAL EFFECTS ON INTELLIGENCE
301
9.6. Training Intelligence
In this section we move from considering the
effects of schooling to the effects of briefer
training programs that are supposed to influ¬
ence general reasoning skills. It is important
to distinguish these programs from programs
that are essentially adjuncts to the formal
education program, in which they teach spe¬
cific topics, for example, reading or mathe¬
matics. Some of the supplementary teach¬
ing programs are good, some are not, and
there is no point in discussing them. Other
commercial programs are marketed as ways
to improve your thinking, which is a rather
nebulous claim. I do not know of a single
one of these programs that has ever offered
acceptable scientific evidence of its effec¬
tiveness. The programs are marketed on the
basis of endorsements by users, rather than
by comparisons between experimental and
control groups. A reviewer with scientific
training will want to scream uncontrolled
placebo effects1.
There is a lively market in coaching pro¬
grams intended to improve students' scores
on socially important tests, such as a college
entrance test. The SAT is a favored target.
These programs do work. It is worth looking
at their claim to have improved intelligence.
In principle, there is another way
to improve intelligence through training.
While we certainly do not know all about
the basic cognitive processes that under¬
lie intelligence, we do know a good deal
(cf. Chapters 6 and 7). Could intelligence
be improved by training basic information¬
processing functions, by analogy to the way
in which athletic performance is improved
by exercise? There has been some progress
along this path.
9.6.1. Coaching Programs that Raise
Test Scores
Suppose that I offered to sell you a training/
coaching program that would improve your
score on test X, where test X is any one of
the commonly used tests related to intelli¬
gence, ranging from the SAT to a progressive
matrix test. Should you buy it?
The question is not fanciful. Several
coaching programs that improve scores on
college entrance and similar examinations
are available. These programs do increase
test scores. But do they increase intelligence?
The distinction between crystallized and
fluid intelligence is relevant. Gc, in its gener¬
alized sense of culturally useful knowledge,
rather than in the more limited sense of
questions about knowledge that find their
way into an intelligence test, is certainly
part of intelligence. Note that the gen¬
eralized definition includes most of what
Robert Sternberg has referred to as “practical
intelligence.”114 Of course, such knowledge
can be acquired by coaching, which then
becomes an extension of formal education.
If you are going to have to take an examina¬
tion in Spanish, it makes sense to be trained
in Spanish, and as you receive this train¬
ing you will probably pick up some nonlin-
guistic but useful information about Spanish
culture and society. Gc will be increased.
It is not surprising that a coaching pro¬
gram would work, and if the price is right,
buy it.
A coaching program might also increase
scores by improving skills that are useful in
test taking, but not useful for much else. An
example is the strategy of improving your
chances on a multiple choice test by rul¬
ing out obviously wrong alternative answers,
and then guessing. It might be rational to
purchase a coaching program that taught
test-specific skills if passing the test were a
goal in itself, but not if you were interested
in improving the socially relevant mental
abilities that the test measured.
Two psychologists at the Educational
Testing Service, Samuel Messick and Ann
Jungeblut, have shown that these distinc¬
tions apply to the coaching programs mar¬
keted as ways to prepare for the SAT.11*
These programs vary from a two- or three-
hour session on “how to take a test” to inten¬
sive tutoring over a period of several months.
Depending on the program, documented
gains on the verbal and mathematical
114 Sternberg, 2003; Sternberg et al., 2000.
u* Messick & Jungeblut, 1981.
302
HUMAN INTELLIGENCE
sections of the SAT ranged from 10 or 20
points per section (.10 in standard devia¬
tion units) to as much as 100 points (1 stan¬
dard deviation unit), which is quite a lot.
There was a correlation of .7 between the
length of the coaching program and the size
of the program’s effect. Not surprisingly,
the shorter coaching programs concentrated
on test-taking skills, while the longer ones
amounted to education, which translates
into a real gain in Gc.
Many people would regard these remarks
as simple statements of the obvious. When
people talk about improving intelligence,
or alternative terms, such as "reasoning” or
"problem solving,” I think they are generally
talking about fluid intelligence - the ability
to solve unexpected or unfamiliar problems.
Can this be done?
Here we bump up against a problem of
definition. It is easy enough to see whether
a training program has developed a person’s
vocabulary, or his or her ability to do specific
types of problem solving, in anything from
carpentry to physics to the law. But how
are we to know whether a person has been
trained to think? A program to teach intelli¬
gence is successful to the extent that it raises
some of the skills that constitute the general
definition of intelligence. In order to mea¬
sure the effect of a program we measure per¬
formance on a test of fluid intelligence. This
alone is not enough. We have to show that
improved test performance has been accom¬
panied by improved performance in socially
relevant tasks outside of the testing context.
Otherwise we are open to the charge that
what has been taught are test-specific skills.
This is a real issue. Not surprisingly, if people
take the same cognitive test twice, they get
better scores the second time. This applies to
tests as different as battery-type intelligence
tests and progressive matrix tests. However,
statistical analyses of the improved scores
show that they are not related to the gen¬
eral reasoning factor (g). Test-specific skills
have been improved by familiarity with the
tests.116
116 te Nijenhuis, van Vianen, & van der Flier, 2007.
What we would like to have is a demon¬
stration that going through some form of
training program that is very different from
a test of g results in improvements on a
g-loaded test. I know of very few such
efforts. The next section discusses two such
attempts, one that succeeded and one that
failed. I then make some speculative remarks
about the sorts of situations that can lead to
success.
9.6.2. Mixed Results from School-based
Programs that Might Improve Gf
The good news for the training of intelli¬
gence comes from an unlikely place - the
Sudan. In 2007 a group of researchers lead
by Paul Irwing, of the University of Manch¬
ester, and including Sudanese colleagues,
conducted a study on the effects of train¬
ing with the abacus, an ancient comput¬
ing device that is still used in the Mideast
and northern Africa. The experimenters
compared schoolchildren’s Raven Progres¬
sive Matrices performance, before and after
the children had either completed a nor¬
mal school curriculum or had had the same
curriculum, plus sixty-eight hours of abacus
instruction, two hours per week for several
months. Both groups showed improvement
on the progressive matrix test, which is not
surprising because it was a second adminis¬
tration, and also because the children were
now almost half a year older. The key point
is whether the experimental groups showed
more improvement than the control.
Figure 9.12 shows that they did. This sug¬
gests that the training program did influence
general reasoning skills.
The negative results come from my own
laboratory. For a number of years my col¬
leagues at the University of Washington and
I, in cooperation with Jim Minstrell, an
award-winning high school science teacher,
worked on the development of educational
programs to improve high school science
instruction, largely by presenting challeng¬
ing problems to be solved, and then empha¬
sizing the reasoning involved.117 In some of
117 Hunt & Minstrell, 1994, 1996.
ENVIRONMENTAL EFFECTS ON INTELLIGENCE
3°3
7 8 9 10 11
Age group in years
ad (start)
■ d (after training)
Figure 9.12. Abacus training improves intelligence test scores. The
difference between the experimental and control groups, in d units,
for Raven Standard Progressive Matrices scores obtained by-
Sudanese children, before and after the experimental group
completed abacus training. The d values within age group are
approximations, based on the data in Irwing et al., 2008, Table 2, on
the assumption that there were approximately equal numbers in the
experimental and control groups.
our studies we had students attempt pro¬
gressive matrix problems before and after
receiving up to a year’s instruction. While
the programs were quite successful in teach¬
ing reasoning about topics in introductory
physics, we did not find any consistent gains
on progressive matrix problems.118
There are so many differences between
elementary schools in the Sudan and high
school students from affluent American
school districts that one hesitates to draw
any conclusions at all. There is also a puz¬
zling aspect to the study. The problems pre¬
sented to the American students stressed
reasoning, not (for the most part) detailed
knowledge of the minutiae of physics. Doing
arithmetic on the abacus does not require a
great deal of problem solving, once you have
grasped the basic principles of the device,
but it does require concentration and an abil¬
ity to hold intermediate calculations in one’s
head. Naively, one might think that solving
physics problems would be more like solving
progressive matrix problems than doing aba¬
cus arithmetic. So why did the results come
out the way they did? Speculation about this
is instructive, because it raises some general
118 Levidow, 1993.
considerations about interpretation of find¬
ings in this area of research.
The authors of the Sudanese study say
that children in both the experimental and
control group were following a standard cur¬
riculum, but provide no details about any
differences in other education-relevant or
environmental experiences that might have
differed between the groups. They did not
indicate whether experimental and control
classes were chosen at random, something
that is essential in evaluating educational
research. My evaluation is that some envi¬
ronmental variable improved reasoning in
the Sudanese children, and that it was prob¬
ably associated with abacus training, but I
would like to have more details.
A second reason that I speculate as being
a cause of the discrepancy in results is that
the participants in the study were operat¬
ing at different levels of general reasoning.
The Sudanese children were operating at the
twelfth percentile of performance, in terms
of the British standardization of the Raven
Progressive Matrices test. A similar statistic
is not available for the American teenagers,
but they were in schools that served mid¬
dle to high SES families, and were voluntar¬
ily taking high school physics, an optional
HUMAN INTELLIGENCE
3°4
course that students consider to be difficult.
It is reasonable to assume that they were in
roughly the upper third of general reasoning
talent.
Recall the consistent finding that we
know quite a few aspects of the environ¬
ment that harm intelligence, and only a
few that make much improvement at the
top. The Sudanese children were probably
operating toward the bottom of the scale
and, compared to the American students,
in quite unfavorable environments. If my
conjecture is correct, their (low) reasoning
scores would be relatively malleable. The
American students were operating at a much
higher level on the reasoning scale, so the
changes that were made to the instruction,
although beneficial in the narrow sense of
teaching physics (the primary purpose of the
intervention), made little if any difference in
terms of general reasoning capabilities.
9.6.3. Training Information-processing
Capacities: Processing Speed and
Automation
Cognitive processing speed refers to the speed
with which the brain can accomplish simple
decision and recognition tasks. Intelligence
theorists have tended to treat processing
speed as a stable trait that has a correlation
with intelligence test scores around .30 in
college students.119 The correlation increases
if we expand our studies throughout the
adult age range, because cognitive slowing
is a general characteristic of aging. What
apparently has not been realized (at least by
most intelligence researchers) is how mal¬
leable processing speed is.
In the 1970s Walter Schneider (now at
the University of Pittsburgh) and Richard
Shiffrin (now at the University of Indiana)
conducted a very important series of exper¬
iments on the influence of practice upon
visual detection. In visual detection tasks an
observer scans an array of stimuli to deter¬
mine whether or not a target stimulus is
present. To illustrate, is the letter K (the
119 Jensen, 2006.
target) present in the following array?
A R V B
H K T P
F W E Q
Schneider and Shiffrin showed that under
certain conditions training can decrease the
time required to detect a familiar target
by a factor of 10, from over 200 millisec¬
onds to as few as 20 or 30 milliseconds.
The training requires hundreds of trials
in which observers always search for the
same target.120 Further research showed that
the phenomenon is not restricted to visual
detection. Similar, somewhat smaller reduc¬
tions in detection time are found when
observers search for exemplars of abstract
categories, such as searching an array of
words for an animal name.121
Schneider and Shiffrin presented their
results as a demonstration of automaticity ,
the principle that if the same task is prac¬
ticed over and over again it becomes very
fast, and not subject to the control of the
relatively slow working memory-attentional
control system. Their explanation has been
verified by imaging studies in which par¬
ticipants have their brains scanned as they
practice tasks to the point of automation.
There is a shift in activity from the forebrain
regions involved in the working memory-
attentional control system to task-specific
regions of the brain.122 This has implications
for higher-order cognition.
Anders Ericsson, a professor at Florida
State University who has conducted system¬
atic studies of expert behavior, has shown
that experts in many fields practice in such
a way as to encourage the automatization
of those components of a complex task that
can be automated.12* This allows the expert
to concentrate his or her attention on those
aspects of a task that require thought. To
take an example from athletics, during a
120 Schneider & Shiffrin, 1977.
121 Fisk & Schneider, 1983.
122 Hill & Schneider, 2006.
125 Ericsson, Krampe, & Tesch-Romer, 1993.
ENVIRONMENTAL EFFECTS ON INTELLIGENCE
match a champion tennis player will concen¬
trate on where to place the ball, not on how
to hit it. That has already been drilled into
the champion’s brain, by hours and hours of
practice.
We may illustrate the same thing with
respect to tasks specifically studied in intel¬
ligence research. I offer an example from
a study done in my own laboratory. Recall
that in the g-VPR model of intelligence124
the ability to mentally rotate a visual percept
“in the mind’s eye” is seen as a basic dimen¬
sion of intelligence. An often-repeated find¬
ing is that mental rotation is faster in men
than in women, and that it deteriorates over
the adult years. With only five days of train¬
ing, one hour a day, we trained middle-aged
women to perform mental rotation tasks as
quickly as they were performed by male
undergraduate students when they walk into
the laboratory. We did not destroy the age
effect, for the undergraduates also got faster
with practice. What we did show is that
mental rotation, the prototypical task for
assessing R in the g-VPR model, is not invari¬
ant over practice.125
Is automation relevant to everyday prob¬
lem solving? The answer to this question is
“yes, almost always a little bit, and some¬
times a lot.” We can think of automation
as an increase in the speed with which
people can recognize that certain responses
are dictated by the situation. Tasks can
be automated if there is a constant map¬
ping between a stimulus and a response; for
instance, you are always supposed to stop at
a red traffic light, and 2 + 4 always makes 6.
Very few of the activities that we call “think¬
ing” can be entirely automated. Almost all
of them contain elements that can be auto¬
mated. Here are two examples.
Mathematics is not the same as arith¬
metic, but mathematicians know their arith¬
metic very, very well. Alexander Aitken, a
leading British mathematician in the mid
twentieth century, was also a formida¬
ble mental calculator.126 Schoolchildren are
124 Johnson & Bouchard, 2005.
125 Berg, Hertzog, & Hunt, 1982.
126 Nickerson, 2010, pp. 159-162.
3°5
drilled until they know their multiplication
table, up to 10 x 10. Several mathematicians
have been reported to know a good part of
the multiplication table up to 100 x 100! It is
important, though, to distinguish between
arithmetical calculation and doing mathe¬
matics. Calculation is a tool for mathemat¬
ics, distinct from any deep understanding of
mathematics. It is hard to imagine a math¬
ematician who could not calculate. On the
other hand, there are numerous cases of cal¬
culating prodigies who had little mathemat¬
ical talent.
What is going on? Mathematicians learn
many arithmetical facts to the point that
they are readily accessible from long-term
memory. As is the case in many other areas
of expertise, these facts are organized into
coherent networks of relationships, so that
they may be easily retrieved. There are spe¬
cialized brain areas associated with calcu¬
lation and simple mathematics, arithmetic
and number representation. In both mathe¬
maticians and calculating prodigies the men¬
tal calculation areas simply run more effi¬
ciently than in most of us. Mathematicians
utilize their efficient mental calculations in
the service of deeper understanding of math¬
ematics. The calculating prodigies stop at
arithmetic.127
And then we have verbal comprehen¬
sion. Here is a fragment of a poem writ¬
ten by Lewis Carroll, the author of Alice in
Wonderland.
“ The time has come/' the walrus said
“ to talk of many things: Of shoes and
ships - and sealing wax, - of cabbages
and kings."
— Lewis Carroll , " The Walrus and
the Carpenter” (18 72)
Figuring out the meaning of this passage
is hard enough without wasting time search¬
ing for the meanings of walrus, shoes, ships,
sealing wax, cabbages, and kings. Automated
word retrieval leaves time for the hard tasks.
An intelligent person will have reduced
the processing time required to retrieve mil¬
lions of items of information that are useful
127 Butterworth, 2006.
306
HUMAN INTELLIGENCE
in everyday life. And how will automation
be achieved? By practice.
9.6.4. Training Working Memory
The working memory-attentional control
complex is a central part of the information¬
processing functions supporting intelli¬
gence. A demonstration that the efficiency
of working memory can be improved is
tantamount to a demonstration that intel¬
ligence can be improved. Why?
Working memory is illustrated by tasks
that require participants to do several things
at once, such as simultaneously monitoring
visual and verbal input streams, or suppress¬
ing information from one input stream while
processing information from another. Such
tasks are not demonstrations of intelligence
in themselves. They are measures of the
functioning of an information-processing
system in the brain, primarily in the frontal
and parietal cortices and in the cingulate cor¬
tex, that provides capacities vital for solving
complex problems.128 The brain has consid¬
erable plasticity; it can reorganize itself as
a result of experience. Two recent studies
have shown that practice on tasks involving
working memory and the control of atten¬
tion will improve performance on intelli¬
gence tests.
One was a study by a joint University of
Michigan-University of Bern (Switzerland)
group that emphasized the storage compo¬
nent of the working memory complex.129
The task used was an n-back task. In this
task a stream of stimuli are presented, one
at a time. The participant indicates when
the current item is a repetition of the item
presented a certain number of items back.
To illustrate, in the stream
XXYZZXYX
the third X is a repetition of the stimu¬
lus presented four items previously. In a
four-back task the participant would be
128 See Chapter 5 for a discussion.
129 Jaeggi et al., 2008.
© ©
Figure 9.13. A cartoon rendition of Posner and
Rothbart's attention-training task. The task is to
press the key showing the direction of the center
fish. The fish on the right and left of the center
may point in the same or the opposite direction
of the center fish.
asked to indicate the repetition. College stu¬
dents attempted a version of the task in
which they were required to monitor two
simultaneously presented streams of stim¬
uli, one presented visually and one presented
aurally. The students practiced for from one
to nineteen days. They also took a variety
of fluid intelligence tests before and after
practicing the working memory tasks. Test
scores increased after practice on the work¬
ing memory tasks, and, most impressively,
the amount of increase was directly related
to the amount of practice.
The second study was done by Michael
Posner and Mary Rothbart of the University
of Oregon.130 Young children were trained
on a visual attention task intended to force
them to focus on one part of a visual
scene while ignoring others. The task itself
is shown in Figure 9.13. Compared to a
control group, the children trained on the
visual attention task showed improved per¬
formance on the Kauffman children's test of
cognitive ability.
The working memory-attention complex
is not the only information-processing func¬
tion that is both relevant to intelligence
and trainable. Processing speed can also be
trained. The mechanism for training is sim¬
ple and ubiquitous in our society: action
video games. It has been shown that ado¬
lescents and young adults who play action
video games perform better on a variety
of tests of general processing speed than
do their contemporaries who do not play
the games. This cannot be entirely a selec¬
tion effect, for it is possible to take people
130 Posner & Rothbart, 2007.
ENVIRONMENTAL EFFECTS ON INTELLIGENCE
who do not normally play games and have
them practice playing; they then show supe¬
rior performance on measures of cogni¬
tive processing speed outside of the gaming
context. 151
Results such as these illustrate how prac¬
tice on tasks that themselves do not evalu¬
ate intelligence, in any usual sense of the
word, can improve performance on the
complex problems presented on intelligence
tests. There is a close analogy to the finding
that intelligence test performance can be
improved by, in Posner and Rothbart's
term, “training the brain/' and the totally
unsurprising, but important, finding that
weight training can improve performance in
athletics.
However, it is important not to overin¬
terpret laboratory studies like these. They
show how an environmental effect on intel¬
ligence could be produced, but they do
not show that variations in intelligence are
produced this way in the world outside
the laboratory. Take, for example, the cor¬
relation between parental SES and chil¬
dren's intelligence. There are differences
in the home environments of children in
high and low SES homes, especially in
the amount of exploratory activity permit¬
ted, parental encouragement of independent
problem solving, and amount (and type) of
TV watching. Do these differences trans¬
late into differences in opportunities to exer¬
cise the working memory-attention control
complex? The question could be answered
by a painstaking cognitive task analysis of
children's environments. To my knowledge,
no such analysis has been conducted.
9.7. The Challenge Hypothesis
The Challenge Hypothesis (Chapter 1) is
a claim that people develop their intelli¬
gence when they rise to meet environmental
challenges. Genetic inheritance establishes a
reaction range, which determines the limits
of development. The level of intelligence a
person achieves within these limits depends
1?1 Dye, Green, & Bavelier, 2009.
3°7
upon both the environment and the way the
person reacts to it. The following principles
apply:
1. The physical environment constrains
the reaction range. The constraints gen¬
erally have the effect of driving intel¬
ligence downward, as in the case of
alcoholism, poor nutrition, and atmo¬
spheric lead. Certain drugs related to
the amphetamines can provide a tempo¬
rary enhancement of alertness, and can
produce transient improvement of some
cognitive functions, but their long-term
effects are not known, and could well
be deleterious. More generally, within
the range of the physical environments
present in post-industrial societies, ben-
efitting from the physical environment
is largely a matter of avoiding things that
will make you stupid.
2. Within a person's reaction range, real¬
ized intelligence is produced by inter¬
acting with the social environment. This
includes the early home environment,
the school, and in adult life, personal
and professional experiences.
3. The extent to which a person benefits
from environmental challenges depends
upon how the person engages with
them. Sternberg152 has identified three
strategies for engagement. Adaption is
accepting the situation and changing
your own behavior to meet it. Decid¬
ing to study for an examination is a pro¬
saic example. Shaping is changing the
situation to adjust to your abilities. This
in itself can be a learning experience.
Selection is finding a situation in which
you can more easily prosper, given your
current talents. When a student drops
a tough class in favor of an easy one,
he or she selects and, as the example
shows, disengages. Each of these strate¬
gies makes sense in an appropriate situ¬
ation, but only adaption and shaping are
likely to lead to expanded intelligence.
Sternberg, 2003b; Sternberg, Grigorenko, & Zhang,
2008.
308
HUMAN INTELLIGENCE
I will now make some frankly speculative
remarks about society and even civilizations.
Social environments vary in the extent
to which they encourage the development
of intelligence, because they vary in the
extent to which they encourage engagement
by adaption and shaping. A society that
endorses the belief that intelligence is fixed
inhibits the development of intelligence,
because it encourages people to avoid chal¬
lenging problems. Look at this rationally. If
you encounter a problem that appears diffi¬
cult, and if you believe that your own abil¬
ities are fixed, then the sensible thing to do
is to select a new environment.
This is particularly the case if the costs
of failure are high. And in the case of
schoolchildren and adolescents, the cost of
failure in front of the peer group is very
high. A person who engages with intellec¬
tually challenging, and hence intelligence-
developing, problems is going to make
errors. The important thing is what hap¬
pens next. There is an apocryphal saying
that engineers analyze an accident in order
to prevent the next one, while lawyers ana¬
lyze an accident in order to find someone to
blame. If we want people to develop intel¬
ligence, we have to encourage them to take
an engineer’s attitude, not a lawyer’s.
Having established this attitude, society
must provide the challenges. This is easy,
providing that a society does not become
fixed in a belief that the way it does things,
now, is the only way to do them. This is
a point where we can be optimistic about
our society. The sociologist Carmi Schooler
has pointed out that the rise in intelligence
test scores across generations has been paral¬
leled by a rise in the complexity of society.1*3
Personal finance provides a good illustration
of Schooler’s point. It used to be that per¬
sonal finance was simple for all but the very
wealthy. If you had the money to buy some¬
thing, you could; if you did not have the
money, you could not. Today we have credit
cards, adjustable rate mortgages, auto loans,
student loans, and what have you. For most
135 Schooler, 1998.
of the citizenry cash is passe, and thinking
about one's personal finances is challenging.
There are people who have argued that the
same could be said of our much-condemned
video games. Some of them contain ele¬
ments of the techniques used to train visual-
spatial reasoning, processing speed, and the
working memory-attention complex!1*4
A dynamic society challenges its mem¬
bers and, as it does so, redefines what it
means to be intelligent. If a society is per¬
vaded by a belief that the old ways are the
best ways (or the only ways), both the devel¬
opment and the definition of intelligence
are constrained. Since history does move on,
the environmentally driven improvement in
intelligence that we are seeing today has
happened before, and probably in a more
extensive way than could be documented by
changes in performance on the intellectual
puzzles we call “intelligence tests.”
The present human species has been
around for roughly 100,000 years. The last
five thousand have seen the development of
ideas that have profoundly changed human
existence. These include the concept of agri¬
culture, literacy, the use of formal laws to
order social conduct, the development of
mathematics and logic, and induction based
on scientific reasoning. These ideas in turn
made possible the spread of technologies
that fostered further new ideas. Agricul¬
ture made economic specialization possible,
which in turn led to urbanization, a need for
formal laws, business transactions, surveying
of land, and construction. Humanity was on
the way toward literacy and mathematics.
People in technologically and intellectu¬
ally demanding cultures respond to the chal¬
lenge, and move toward the top of their
reaction ranges - at least with respect to
those aspects of cognition that their cul¬
ture sees as important. The Sumerians were
smarter than the nomads on the steppes, and
the Roman soldiers who manned Hadrian's
Wall were smarter than the Piets and Celts
outside it.
134 See Greenfield, 1998, for some early examples, espe¬
cially in the spatial-visual realm, and Dye, Green, &
Bavelier, 2009, for more recent work.
ENVIRONMENTAL EFFECTS ON INTELLIGENCE
Nobody, and no culture, thinks of
everything. Charles Murray has coined the
elegant term metainvention to describe ideas
that have profound implications for society,
such as logic and the rule of law.135 He points
out that virtually all metainventions have
arisen along the Eurasian land mass, primar¬
ily in northern Europe, with a “recent” (since
1600 CE) rise in the Americas, following the
European migrations to the New World.
Why?
Galton and his colleagues in Victorian
England would have said that this is because
the northern Europeans are genetically
superior, with respect to intelligence. There
are scientists today who would agree with
him, a topic that will be taken up in Chap¬
ter 11. The genetic hypothesis cannot be
refuted on the basis of the geographic dis¬
tribution of intellectual contributions. But
there is an environmental explanation.
The ecologist Jared Diamond136 has
argued that great ideas and cultural inno¬
vations arise when societies meet, both
to exchange and to challenge ideas. Dia¬
mond further believed that this psycho¬
logical/sociological process interacted with
geography. The East-West orientation of
Eurasia permits migration, travel, and trade
literally from the Atlantic to the Pacific.
Such movements produce exchanges and
development of ideas. The Americas and
Africa are oriented North-South, so that
ecological barriers such as the jungles of
Panama and the Sahara Desert discourage
travel along the major axes of the conti¬
nents. A particularly telling example is in
the New Guinea highlands, where Melane¬
sian peoples developed agriculture indepen¬
dently of its appearance in the Middle East,
but otherwise remained in the Neolithic Era.
Diamond argues that this was not because
of a genetic difference between Asians and
Melanesians; it was because of the isolation
of the New Guinea highlands.
Geographic isolation is not the only way
to prevent the development of intelligence
by engagement; social isolation works as
135 Murray, 2003.
136 Diamond, 1997.
3°9
well. The Chinese Empire in many ways
failed to participate in the avalanche of ideas
that started with the European Renaissance
and the Age of Exploration. Why? At least
in part because the conservative philosophy
of Imperial China left the Chinese uninter¬
ested in other cultures.
At both levels, the society and the indi¬
vidual, intelligence increases when there is
a combination of a challenge and a willing¬
ness to rise to it. Humans will never produce
a society that makes intellectual demands
beyond the human potential - by defini¬
tion. How close individuals come to reach¬
ing their cognitive potential depends on how
much thought their society demands from
them.
9.8. Summary: What Produces
the Cohort Effect?
Just as nonzero heritability coefficients show
that genetic inheritance is an important
determiner of intelligence, the rise in test
scores throughout the twentieth century
shows that environmental effects can be very
powerful. These demonstrations, alone, do
not tell us how genetic and environmental
variables act upon intelligence. In Chapter 8
we saw that great progress has been made
toward finding out which genes reduce
human intelligence markedly, but that it has
been much harder to find the genes that pro¬
duce variation in the normal range.
Much the same can be said of the phys¬
ical environment. We know that we should
avoid poor nutrition, atmospheric lead, and
other environmental hazards. Very little
progress has been made toward identifying
the positive aspects of the environment that
further intellectual development, within the
range of environments present in the indus¬
trially developed countries.
Schooling is clearly important. Over the
course of the twentieth century the devel¬
oped countries reached a goal of almost uni¬
versal education. In addition to increasing
the level of education of the average person,
the developed societies made a huge effort
to reduce the number of school drop-outs.
310
HUMAN INTELLIGENCE
This is almost certainly one reason that the
cohort effect is largely driven by increases in
the absolute level of test scores in the lower
range of cognitive ability.
The effects of changes outside of formal
schooling are harder to establish. While
some very intense pre-school programs,
such as the ABCDerian effort, have shown
promising long-term results, projects of this
intensity have affected only a few hun¬
dred students. This is not nearly enough to
influence population-level IQ scores. More
widespread programs, such as the American
Head Start program, are probably not inten¬
sive enough to make permanent changes in
cognitive capabilities.
Family sizes have dropped precipi¬
tously in the developed countries. Simi¬
lar reductions in family size are now being
seen worldwide. Even after allowance for
parental intelligence is made, smaller fami¬
lies produce somewhat more intelligent chil¬
dren. However, these effects are too small to
account entirely for the cohort effect.
Intelligence is statistically associated with
a tendency to become intellectually engaged
with challenging tasks. Although correla¬
tion does not mean causality, it is at least
arguable that those people who engage in
intellectually challenging activities improve
their cognitive capacities - their intelligence
in a far larger and more important sense
than in the narrow sense of improving intel¬
ligence test scores. A plausible argument can
also be made that throughout the twentieth
century there was an increase in the cogni¬
tive complexity of the environment. People
really were getting brighter, at least in the
sense of greater acquisition of those cogni¬
tive skills evaluated by the tests.
Are these intelligent people benefitting
from their capabilities? In the next chapter
we consider what intelligence is worth in the
post-industrial society.
CHAPTER 10
What Use Is Intelligence?
As of the end of the twentieth century, the
United States is run by rules that are
congenial to people with high IQ and that
make life more difficult for everyone else.
Herrnstein & Murray, 1994, p. 541
The quotation from The Bell Curve: Intelli¬
gence and Class Structure in American Life , is
a pretty strong statement about the impor¬
tance of intelligence. When Herrnstein and
Murray made it they were attacked as eli¬
tist and antidemocractic. Other people, with
impeccable democratic credentials, had said
similar things in a less contentious way. Just
a few years before Herrnstein and Murray
wrote, Robert Reich, a sociologist who had
served as Secretary of Labor in the Clin¬
ton administration, wrote that work has
shifted from emphasizing the manipulation
of objects to the manipulation of abstract
ideas, varying from programming a robot
to analyzing a financial system.1 It follows
that skill in manipulating abstract concepts,
intelligence, has become progressively more
1 Reich, 1991.
valuable over time. To what extent do cog¬
nitive tests predict such skill?
10.1. Problems in Investigating the
Relationship between Intelligence
and Success
This chapter examines the relation between
intelligence and success in three broad
regions; academics, the workplace, and per¬
sonal life. These studies are not easy to
do, for several reasons. First, we have to
specify what we mean by success in each
arena. Next, we have to select quantita¬
tive measures of success. Observable mea¬
sures, such as grade point average or money
earned, are frequently only partially satisfac¬
tory measures for our criteria. They often
have undesirable statistical and measure¬
ment properties that hinder analysis and
interpretation. Finally, there is the impor¬
tant problem of generality. We cannot study
the totality of academia, the workplace, or,
certainly, personal life. We have to study
slices of them, where the necessary measures
can be obtained. These slices are almost
312
HUMAN INTELLIGENCE
never random samples of the arenas, and
in only a few cases can we obtain the ideal
experimental group-control group contrast.
As is true in other areas of research on
intelligence, we can learn something from
imperfect studies. We just have to keep the
imperfections in mind when we consider
what has been learned. The rest of this sec¬
tion describes some of these imperfections.
Do not lose sight of the magnificence of the
forest because the trees have woodpecker
holes in them! Quite a lot has been learned.
lo.i.i. The Conceptual Criterion Problem
The biggest problem is defining success. In
the academic arena a student is successful if
he or she has learned. The commonest mea¬
sure of academic success is a person’s grade
point average (GPA) across classes. How¬
ever, grade point averages are not compara¬
ble across classes or institutions. A student
with a GPA of 3.5 in English classes in a
community college is not necessarily a bet¬
ter student than one with a GPA of 3.1 in
Physics at Stanford. Merging gross measures
of learning, such as GPA, across subjects or
across schools introduces unwanted sources
of variance. This will make the intelligence-
GPA relation appear to be smaller than it
is. But measuring the relation in one class
or institution raises question about how the
finding can be generalized.
An alternative measure of success is grad¬
uation or, in the K-12 system, its inverse,
dropping out. Once again we have noncom¬
parability across institutions; without nam¬
ing names, not all our high schools, colleges,
and universities are equivalent. Americans
keep school records by district or state, not
by a national register. If a student disappears
from a K-12 system or fails to complete a
postsecondary program, there is no record
of where that student went. They may have
dropped out, or they may have enrolled at
another educational institution.
It is even harder to define success in the
workplace. Within an industry or occupa¬
tion income partially captures the idea of
success, but incomes across occupations are
hard to compare. Incomes are also often
determined by variables unrelated to intel¬
ligence, such as seniority of employment.
Some of our larger companies do keep
records of periodic evaluations of employee
performance, most commonly supervisors’
ratings. Ratings are not reliable unless the
raters are trained and the criteria for rating
have been agreed upon. Objective measures
of employee output are often hard to come
by and generally capture only a part of a
person’s job. COSTCO, a giant warehouse
sales company, tracks the number of check¬
outs per hour that each of their check-out
clerks handles. It does not directly measure
things like customers’ reactions to a clerk’s
manner.
Defining success in life is even harder.
We can measure extreme social adjustment,
which can vary from achieving a civic prize
to going to jail, but most people do neither.
Success in life is a multifaceted thing. Infor¬
mative studies have been conducted of the
relation between intelligence and particular
aspects of life success, such as health, but
trying to relate intelligence, or virtually any
other trait, to such a nebulous thing as ‘life
success” is probably not a useful exercise.
Once we have defined our criteria we
face the problem of actually getting the
data. Several strategies have been followed.
One is to conduct an experimental study, in
which the investigator obtains measures of
both intelligence and success from a selected
set of participants. To take an example,
one study related intelligence test scores
to success as a race track gambler.2 Such
studies tend to be fairly small and to deal
with unique situations. Because they are
small, they can detect only large relation¬
ships. (Technically, they have low statisti¬
cal power.) This brings us to a discussion of
statistical issues.
10.1.2. The Statistical Problems
We measure the extent to which intelligence
is related to some index of success by cal¬
culating predictive validity, which is defined
as the correlation between a measure of
2 Ceci & Liker, 1986.
WHAT USE IS INTELLIGENCE?
intelligence and the criterion measure. The
process is sensitive to three statistical issues:
reliability, range restriction, and gener¬
alization. In order to understand them
we need a brief digression into statistical
reasoning.
THE RELIABILITY ISSUE
Any measurement contains two elements,
a “true value” and a residual term. While
the residual term is frequently referred to as
“error,” it is not necessarily error in the sense
of a mistake. It refers to the sum of all influ¬
ences on the measured variable that are sta¬
tistically independent of the true value. To
take an example, consider the way in which
weight is measured during the typical annual
physical. Examinees are told to take off their
shoes and stand on a scale. Measured weight
is then shown on the scale. The measured
weight has the following components:
Measured weight = actual body weight +
( weight of clothes + scale bias),
where scale bias refers to any tendency of
the scale to weigh high or low. The terms
in parentheses, here weight of clothes and
scale bias, are residual effects, uncorrelated
with the examinee’s actual weight. If an
examinee were to be weighed on a different
scale, wearing different clothes, measured
weight might change even though actual
body weight remained the same. Measured
weight is said to be reliable to the extent that
the same measure is obtained across compa¬
rable conditions. This reasoning applies to
intelligence testing.
An intelligence test score x is determined
by the examinee’s “real” intelligence and a
residual term that is unique to the exami¬
nation of that person at that time. Exactly
the same thing can be said of an academic
grade, y. The grade is determined in part by
what the student really knows about, say,
English Literature and in part by a residual
term unique to the examination and the per¬
son. Symbolically,
0°0
where the subscript t stands for “true” and
e denotes the residual term. Now define the
reliability of an intelligence test or a grade as
the correlation between two measures, each
assumed to be equally good, taken on the
same person. Examples would be the corre¬
lation between two equivalent forms of the
SAT, or the correlation between the grades
assigned to the same set of English Litera¬
ture examinations by two equally qualified
graders.
What we can observe is the correla¬
tion between test scores and grades, r xy.
What we want to know is the correlation
between intelligence and academic achieve¬
ment, rXtVy . This is
where r** and r yy are the reliability corre¬
lations for the intelligence test, x, and the
academic measure, y. The correlation r ^ is
sometimes referred to as the “true” correla¬
tion.
As reliability coefficients range between
o and 1, the denominator, *Jrxxryy, will
also range between zero and one. There¬
fore, the correlation between the “true” vari¬
ables, r^y,, will be at least as big, and gen¬
erally larger, than the correlation between
the observed variables, rxy. Corrections for
unreliability have to be treated with caution,
as the reliabilities are themselves estimates,
and if they are too low the estimated true
correlation can exceed one, which is obvi¬
ously not correct.
In both the academic and industrial
cases the difference between the observed
correlation and the (estimated) true correla¬
tion can be substantial. Professionally devel¬
oped intelligence tests generally have relia¬
bilities of .85 or above, but this is because
of a great deal of careful item selection
(Chapter 2). Large-scale academic achieve¬
ment tests, such as those used in the United
States to assess educational progress on
a statewide basis,3 have similar reliability
3 As of 2009 such tests were required by federal law -
the No Child Left Behind Act.
x = xt + ex; y = yt + ey,
HUMAN INTELLIGENCE
3M
coefficients. Within-class, teacher-assigned
grades are quite another matter. I know of
one case where thirty essays were graded,
independently, by two university profes¬
sors. The correlation between the two sets
of grades was .3I4 Fortunately, this is an
extreme. In most cases grades have reliabili¬
ties in the .6 to .8 range. This means that if a
typical study of the relation between intel¬
ligence and grades within a class produces a
value of r, that value should be multiplied
by approximately 1.38 to estimate the true
correlation between intelligence and aca¬
demic achievement.5 This substantial cor¬
rection applies to studies of grades within a
class. As the GPA is an average across classes,
the reliability of the GPA is much higher
than the reliability of a grade within a class,
so the correction would be smaller.
Probably the most commonly used crite¬
rion for achievement in industrial settings is
a supervisors rating of performance. Unless
these ratings are the result of carefully struc¬
tured evaluations they are likely to have reli¬
abilities in the .6 range or lower, consider¬
ably lower than the typical reliabilities of
cognitive tests.
RESTRICTION OF RANGE
In most studies the variability of intelli¬
gence in the group actually studied will be
smaller than the range in the population to
which we wish to generalize. The problem
of estimating intelligence-grade relations in
elementary schools (K-5) illustrates the sit¬
uation. Elementary schools generally draw
students from the neighborhood immedi¬
ately around them. In most industrially
developed countries neighborhoods tend to
have distinct socioeconomic and sometimes
demographic characteristics. Therefore, the
student demographics in a single elementary
school will be more homogeneous than in
the district or state. We say that the scores
are subject to range restriction. In the case
4 The statement is based on personal observation.
5 This conclusion follows from the following argu¬
ment. Assume that the reliability of the intelligence
test is .88 and the reliability of the grade is .6. By
application of equation 10.2, the observed correla¬
tion should be multiplied by 1/1/528 = 1.38.
of the elementary schools, we would expect
there to be less variability in a measure of
intelligence within a school than across a
state.
Range restriction influences the correla¬
tion coefficient. Let crs be the standard devi¬
ation of observed scores in the sample, and
crp be the standard deviation in the pop¬
ulation (in the school and in the district,
in the elementary school illustration). The
relation between the observed correlation in
the sample, rS) and the correlation to be esti¬
mated in the population, rp, is
In this equation rp is greater than rs if
o p > crs. Note that this corrects only for
attenuation on one of the two variables,
either the intelligence variable or the crite¬
rion variable. Correction on both variables
is also possible and often reasonable. For
instance, suppose that a study were done in
which we observed the correlation between
intelligence test scores and scores on an
achievement test in a school, and we wanted
to estimate the correlation in the state. It
would be appropriate to correct for range
restriction on both the intelligence test and
the achievement test.
Selection restriction is an important special
case of range restriction. Selection restric¬
tion occurs whenever an applicant popula¬
tion takes a predictor test, here some sort
of intelligence test, as part of application
for a job or educational opportunity. All
applicants above a given cut score are then
accepted, and their performance on the job
or in the school is recorded. For exam¬
ple, suppose a university uses an entrance
examination, and admits the top 50% of the
applicants. In order to validate the entrance
examination, university officials would want
to know if it was a good predictor of the
grades that an applicant would obtain. How¬
ever, grades are available only for the admit¬
ted students. Since, obviously, the top 50%
of the applicants will have less variation in
WHAT USE IS INTELLIGENCE?
examination scores than the entire group of
applicants, the correlation between exam¬
ination scores and grades in the applicant
population can be estimated by comput¬
ing the examination-grades correlation in
the admitted group, and then correcting for
range restriction.
Corrections for restriction in range can be
substantial. A reasonable value for the hypo¬
thetical university example is crs = .6 crp.6
An observed correlation of .33 in the selected
students would be corrected to .50 for the
applicant population. As correlations are
often squared and reinterpreted as repre¬
senting "percentage of variance accounted
for,” this would change r2 from 11% to 25%,
a substantial change.
Because corrections for reliability and
range restriction can be considerable, know¬
ing when to use them is important. Here are
some general rules.
1. Correction for reliability is appropri¬
ate when one’s interest is in theoreti¬
cal constructs underlying measures - for
instance, whether intelligence as a con¬
cept is related to academic ability as a
concept. The correction is not appro¬
priate when one’s interest is in whether
one set of scores predicts the value of
another set of scores - for instance, if
you wanted to know whether the SAT
predicts first-year college GPA.
2. Correction for selection restriction
should be done whenever the purpose
of the study is to determine the valid¬
ity of a predictor, such as an entrance or
hiring examination.
3 . Correcting for range restriction is appro¬
priate when the available observations
are known to be a nonrandom sample
in which scores are less variable than
they are in the population. The case of
using observations within a single school
to estimate a population in the district
is an example. However, in such cases
6 In the case of selection restriction the percentage
of applicants accepted determines the relationship
between the variances in the sample and the popu¬
lation. In other cases of range restriction this has to
be estimated.
315
correcting for range restriction will be
possible only if an estimate of the pop¬
ulation standard deviation is available.
4. If the sample is a true random sample of
the population, the correction for range
restriction should not be used.
5. Any correction for range restriction car¬
ries with it the assumption that the same
(linear] relation holds between scores
in the sample and in the population.
This is not a trivial assumption. To
take one example, there is evidence that
the relation between IQ test scores and
adult age is nonlinear. Scores decline
more sharply with age beyond sixty
than before. Therefore, it would not be
appropriate to apply range restriction to
estimate the age-IQ relation in adults
from a sample of adults age sixty and
older.
Rule 5 leads to a discussion of our last
statistical issue, power.
Statistical power. To explain these issues,
we need a bit of notation and a review of
introductory statistics.
By tradition, scientific results are said to
be "statistically significant” if they would be
obtained by chance only in fewer than 1 out
of 20 studies [p < .05] or fewer than 1 out
of 100 studies ( p < .01), on the assumption
that the variables being studied in a sam¬
ple are actually unrelated in the population
(the “null hypothesis”). In research on intel¬
ligence, “unrelated” means that in the pop¬
ulation there is no correlation between the
predictor (an intelligence test score) and the
criterion, rp = o. However, rp cannot be
observed directly. Instead it is estimated by
an observed correlation, rSf in a sample of N
observations.
Assuming that the sample can be
regarded as being chosen randomly from the
population, there will be some critical value
of the observed correlation, r*, such that
if the observed correlation, rS) exceeds that
value (ry > r*) we reject the null hypothesis
that rp = o at some level, p, where p refers to
the probability of observing rs > r* if the null
hypothesis is true. The value of r* increases
if we lower the significance level (typically
3l6
HUMAN INTELLIGENCE
from p = .05 to p = .01) and decreases as
the size of the sample, N, increases. To take
some examples, at the p < .05 level the criti¬
cal value, r*, is .36 for a study with 30 obser¬
vations (N = 30], and .20 for a study with
N = 100. At the p < .01 level the values are
.46 and .26.
This much is taught in elementary statis¬
tics. The second point is taught but often
not stressed. Suppose that the sample cor¬
relation is less than the critical value, rs <
r*. This means that we cannot reject the null
hypothesis. “Not rejecting” is not the same
as accepting. What we have is what, in law,
would be called a verdict of “not proven.”7
Suppose that the population correlation
is some value other than zero. (For sim¬
plicity, consider only positive values.) There
would still be some probability that the sam¬
ple correlation fell below the critical value -
that we observe rs < r* even though rp >
o. This probability depends upon what the
population value is, so the probability has
to be specified given a population value and
the size of the study, Pr[rs < r*\rp = k, N).
The power of a study is the complement of
this,
Power[rp — k, N)
= Pr(rs > r*\rp = k, N). (10.4)
In words, this is the probability that a sample
of size N, drawn from a population in which
the population correlation has value k, will
have a sample correlation above the critical
value. Going back to the earlier example,
suppose that the population correlation is
.50. If we set the significance level atp < .05,
the power of a study with a sample size of 25
is .84. This means that 16 out of 100 samples
will not reach a value reliably greater than
zero even though the population correlation
is a substantial .50.
Power increases with sample size. In the
example just given, if the sample is increased
to 100, the power is greater than .995.
The power problem becomes critical
when it is combined with the problem of
7 Such verdicts are not allowed in US courts, but they
are allowed in some countries.
criterion reliability. Grades within a class
and employer rating systems will often have
a reliability of around .60, and intelligence
tests will have a reliability of about .85. Sup¬
pose that the true correlation between intel¬
ligence and academic ability, the hypotheti¬
cal variables underlying these measures, is
.50 in the population. A bit of algebraic
manipulation of equation 10.3 will show
that the expected population correlation
between test scores and grades is .26 after
correction for attenuation. Setting the sig¬
nificance level at .05, the power of a study
with 25 is participants is approximately .36.8
About two out of three studies of this size
would not provide strong enough evidence
to reject the null hypothesis that rp = o,
even though it is false. If the sample size
were to be increased to 100, power would
increase to about .75. In this case failure to
reach statistical significance would be rea¬
sonable evidence against the hypothesis that
a “true score” value of rp was .50 or larger.
What these examples show is that power
is produced by an interaction between the
reliability of the measures and the size of
the study. This interaction has to be taken
into account in evaluating null results. Do
people actually fail to do this? The answer
is, stunningly, “yes.” Panel 10.1 presents the
case of a widely cited study in which no con¬
sideration was given to these issues.
10.1.3. Drawing Conclusions in the Face
of Statistical Uncertainties
Given all these problems, can any conclu¬
sions at all be drawn? The answer is “yes,”
but only after careful consideration.
When evaluating empirical results we
have to consider which statistic is appropri¬
ate. Are we interested in the observed cor¬
relation, or should the correlation be cor¬
rected for reliability and/or restriction in
range? The rules given in the previous sec¬
tion apply.
We must be aware of power consider¬
ations. We need to be especially wary of
8 Power estimates are based on Table 3.3.2 in Cohen,
1988.
Panel 10.1. A Day at the Races:
A Failure to Consider Power
and Reliability
In 1986 two Cornell University psycholo¬
gists, Stephen Ceci and J. K. Liker, pub¬
lished an eye-catching article entitled "A
Day at the Races.”* They reported a four-
year study of the expertise of a group of
thirty habitual bettors on harness racing.
Ceci and Liker did not study the accu¬
racy with which these bettors predicted
winners because, as they said and as
many horse racing fans know, the winners
are often determined by unpredictable
events. Instead they studied the accuracy
with which the bettors were able to pre¬
dict the favorite and top three favorites at
post time (the start of the race), given the
extensive information about each horse
that was contained in the daily racing
form, which is available to bettors prior
to a race. Although secondary references
often misinterpret this, what Ceci and
Liker actually studied was their partici¬
pants' ability to predict how other bettors
would place their bets, not which horse
would win.
Ceci and Liker found that the par¬
ticipants’ decision was a mathematically
complex function of the information con¬
tained on the racing form, and that the
accuracy of the predictions had a corre¬
lation of —.03, essentially zero, with the
participants’ IQs scores on the short form
of the WAIS.* Ceci and Liker drew the
following strong conclusion:
(a) IQ is unrelated to performance at
the racetrack but, more important (b)
IQ is unrelated to real-world forms of
cognitive complexity.
Ceci Liker , 1986, p. 255
These are strong words indeed. The
null finding was claimed to be reliable,
and the task, something that is related to
but not the same as picking the winners in
a race, was unhesitatingly generalized to
the universe of complex tasks. Nowhere
in the article was there any mention of
reliability or power.
Douglas Detterman and his colleague
Kathleen Spry wrote a detailed critique
of the Ceci and Liker study.1 Among
other things, they observed that Ceci and
Liker’s criterion, the ability to predict the
odds at post time, had a reliability of at
most .41. What does this mean? Suppose
that the correlation between the underly¬
ing abilities, intelligence and skill at set¬
ting the odds, is 1. In terms of the text,
rp = 1. The reliability of the short form
of the WAIS is known to be .85. There¬
fore, the expected value of the correlation
in the sample would be .85 x .41 = .35. If
N = 30, the power of the Ceci and Liker
study would be approximately .5, which
means that a study like theirs should fail
to reach the convention .03 level of statis¬
tical significance five out of ten times even
if the underlying correlation was one.
Of course, nobody thinks that the cor¬
relation between intelligence and race
track betting is identically one. Based on
meta-analysis, a widely quoted estimate
of the correlation between intelligence
and performance on a cognitively com¬
plex task is rp = .5.** To be generous,
increase this to .6. Then, solely on the
basis of reliability, the expected sample
correlation would be rs = .21. Using this
estimation the power of the Ceci and
Liker study was .20; studies like theirs
should fail to reach the .05 level of sig¬
nificance four out of five times.
The Ceci and Liker study presents
us with good news and bad news. The
good news is that when a published
study contains major flaws, other scien¬
tists point out the errors. The bad news
is that almost no one notices the correc¬
tion. The Ceci and Liker study has been
cited ninety times, as evidence that intel¬
ligence, as measured by the IQ tests, is
not related to real-world cognition. The
Detterman and Spry study has been cited
seven times.**
* Ceci & Liker, 1986.
T This figure is a correction to the original value,
provided in Ceci & Liker, 1987.
’ Detterman & Spry, 1988. This much-neglected
article contains several other strong criticisms of
the Ceci and Liker work.
5 Schmidt & Hunter, 1998.
** Data collected from an ISI Web of Knowledge
citation search, July 2, 2009.
3i8
HUMAN INTELLIGENCE
concluding that there is no relation between
intelligence and some other variable when
the study involved is small or uses a measure
of unproven reliability.
One way to address the power issue is
to do studies with a very large number of
participants. The extreme case is to uti¬
lize large surveys, such as the Department
of Labor longitudinal studies of US citi¬
zens described in panel 9.9. Sometimes sur¬
veys are “constructed” by analyzing records
of intelligence, health, educational accom¬
plishment, and occupational status that have
been collected for other purposes.
Obviously, though, the larger the study
and the more time required of the partic¬
ipants, the more costly the experiment or
survey. In many cases the investigator has
to accept less-than-ideal measures, such as
using a brief vocabulary test as a proxy for a
measurement of intelligence, or using place
of residence as the sole measure of a par¬
ticipant's socioeconomic status. Such com¬
promises are not fatal errors; they are things
that have to be considered when evaluating
results.
Another way to address the power prob¬
lem is to use a statistical technique called
meta-analysis to draw conclusions from mul¬
tiple studies. 9 Special statistical methods are
used to identify trends that may not be clear
from focusing on the details of any one
study. This technique can be quite reveal¬
ing. However, there are reservations.
As is the case for any statistical technique,
generalizations based upon any unjustifiable
assumptions of random sampling are sus¬
pect. For example, many studies have been
conducted of the relation between scores on
college entrance examinations and student
grades. Meta-analysis can, and has, been
applied to these studies. The participating
institutions tend to be the larger institutions,
with budgets sufficient to support internal
research organizations. Therefore, the result
of a meta-analysis of such studies is a useful
descriptive statement, but any appeal based
upon a claim of random selection of institu¬
tions is questionable.
9 Hunter & Schmidt, 1990.
The individual studies reviewed in a
meta-analysis will, inevitably, vary in the
quality of the measurements taken and the
procedures used. These considerations are
judgments that have to be made by con¬
sidering the details of each study. They do
not lend themselves to statistical treatment.
All reviewers will agree, for instance, on the
number of participants in a study, and the
effect this has on statistical power. They may
not agree on the appropriateness of the mea¬
sures used, or the way in which the mea¬
sures were taken. Some meta-analyses have
attempted to deal with this problem by clas¬
sifying studies by their perceived quality,
and then analyzing high- and low-quality
studies separately, to see if this makes any
difference. A finding that only appeared
in low quality studies would certainly be
treated with suspicion.
10.1.4. Problems Related to
Research Design
The final problems to be considered have to
do with research design, rather than statis¬
tics and measurement.
The ideal research design is a prospective
study, in which the investigator obtains data
on the intelligence of people at some point in
their lives, ideally before they enter an aca¬
demic program or the workforce, and then
determines how well they succeed. This is
by far the easiest kind of study to interpret.
However, it is possible only if the investiga¬
tor has some way of testing a large number
of people, and then following them for a rea¬
sonably long period. There are a few studies
that have done this. The Seattle Longitu¬
dinal Study10 (panel 9.2] and the National
Longitudinal Studies (panel 9.9) are impor¬
tant examples. However, such studies are
expensive, and so are few and far between.
Prospective studies can sometimes be
conducted by examination of government
records. Studies of this sort have been car¬
ried out in those European countries in
which eighteen- to twenty-year-old men
have to register and be tested for potential
10 Schaie, 2005.
WHAT USE IS INTELLIGENCE?
military enlistment. (As far as I know, Israel
is the only country that requires registration
for both men and women.) Valuable infor¬
mation can be gained if some of the reg¬
istrants can be reinterviewed later in life,
to determine how well they have fared.
In some countries this can be done with¬
out actually interviewing the individuals,
because the government keeps extensive
records of the health, education, and income
of its citizens. Legal and ethical issues con¬
cerning access to such data have to be
resolved, but the important point is that the
studies often can be done.
The alternative to a prospective study of
the relation between intelligence and suc¬
cess is a retrospective study. In a retrospec¬
tive study a group of people are identi¬
fied who have, or do not have, varying
degrees of social success. The investigator
then attempts to determine their intelli¬
gence, either by direct testing or by exam¬
ination of relevant records. Studies of emi¬
nence or genius often fall into this category.
The investigator identifies a group of indi¬
viduals who meet some criterion for accom¬
plishment and then tries to identify the com¬
mon characteristics of the group. Possibly
one of the most ambitious of these stud¬
ies was Simonton's determination of the
correlation between a measure of intellec¬
tual capacity, reconstructed from historical
records, and historians’ ratings of the perfor¬
mance of the forty-two US presidents, from
George Washington through George Bush.11
The correlation was .56.
Studies of the relation between intelli¬
gence and success are prone to collinear-
ity problems. To illustrate, intelligence test
scores during adolescence are positively cor¬
related with subsequent income.12 Does this
mean that high intelligence causes a rise in
income? Perhaps. But it is also true that
children's test scores are positively corre¬
lated with parental socioeconomic status,
although the correlation (^.40) is not as high
as many people assume. Is current income
due to intelligence, or is it a legacy of the
11 Simonton, 2006.
12 Herrnstein and Murray, 1994.
319
privilege of having come from a wealthy (or
poor) background, with concomitant oppor¬
tunities (or lack of opportunities) to get a
foothold on the economic ladder? Or both?
10.2. The Relation between Intelligence
and Academic Achievement
Binet’s motivation for constructing the orig¬
inal intelligence test was to identify children
who were at risk for failing in the standard
academic system. Subsequent test develop¬
ers generalized the goal to predicting degrees
of success at all levels of education. How
well has this worked?
10.2.1. Intelligence in the K-12 System
In 1972 the American clinical psychologist
Joseph Matarazzo reviewed the evidence,
and concluded that IQ, as measured by the
Wechsler tests, was a good predictor of high
school graduation.13 In 1994 Herrnstein and
Murray addressed the same question using
the AFQT and the NLSY79 data. Figure 10.1
shows their results for graduation rates in
the 1980-85 period. The finding is clearly
robust. Different tests were used at differ¬
ent times, and with different definitions of
“failure to graduate.” Nevertheless, both
studies found the same positive relation
between test scores and probability of
graduation.
Matarazzo also said that, based on “thou¬
sands” of studies, it had been shown that
intelligence test scores correlate .50 with
grades in the K-12 system.14 This estimate
has been widely accepted by subsequent
reviewers.15 Later reviewers do qualify the
estimate, by saying that the correlations tend
to be higher in elementary than in mid¬
dle school, and drop to perhaps .40 in high
school. As psychological studies are noto¬
rious for failures to replicate findings, the
agreement among reviewers, over a consid¬
erable time span, is reassuring.
13 Matarazzo, 1972.
14 Matarazzo, 1972, p. 283.
15 Brody, 1992, pp. 251-254; Jensen, 1998; Macintosh,
1998, Chapter 2; Neisser et al., 1996.
320
HUMAN INTELLIGENCE
Figure 10.1. Percentage of White young adults in the NLSY79
survey who did not complete high school, plotted as a function of
their percentile scores on the AFQT. Data from Herrnstein &
Murray, 1994.
The .50 figure applies to measures of
grades computed across classes, elementary,
middle, or high school GPA. If the corre¬
lation is calculated for scores within a sin¬
gle class, it will drop due to range restric¬
tion and, often, due to the lowered reli¬
ability of locally produced tests, A study
by researchers at the University of Penn¬
sylvania found a correlation of just slightly
over .30 between Otis-Lennon test scores
(see Chapter 2) and academic achievement
on tests at the end of the eighth grade.16
This study was done in a "magnet” high
school where the students had already been
selected on the basis of test scores and previ¬
ous grades, so range restriction was certainly
a factor. Other studies confined to a single
school or class have found correlations of
about .5 between test scores and grade point
averages in the early primary grades and in
high school.17
Macintosh1" has observed that although
restriction of range is frequently appealed
to as a mechanism that should reduce test-
achievement correlations, the effect has
never actually been observed, at least in
studies of the K-12 system. Two large Euro¬
pean studies come close to addressing Mac¬
intosh’s concern.
16 Duckworth & Seligman, 2005.
17 Kaplan, 1996; Zwick & Green, 2007.
18 Macintosh, 1998.
The English system of education is much
more centralized than the American. One
year something over 78,000 students were
given the Cognitive Abilities Test (CAT,
described in Chapter 2). The test takers
included almost all the eleven-year-olds in
England, so range restriction is not rele¬
vant. At age sixteen all English students take
nationwide examinations in a variety of sub¬
jects. The national examinations are subject
to much more careful psychometric evalua¬
tion than is typically the case for locally gen¬
erated (and certainly for teacher-generated)
examinations, so reliability of the criterion
variable was not a major concern.
Ian Deary and his colleagues19 extracted
a general intelligence (g) factor from the
CAT scores and a general academic achieve¬
ment factor, which I will call a , from the
scholastic examination scores. The correla¬
tion between the two was rga — .81. This
is a very high value. There was substan¬
tial variation between associations with the
g factor and educational accomplishment
within topics. Correlations ranged from a
high of .77 for mathematics to .43 for Art
and Design. In general, the topics usu¬
ally considered the academic core courses -
the Humanities, Mathematics, and the Sci¬
ences - had correlations in the .50-. 75 range,
while “practical” topics, such as Art and
19 Deary, Strand, et al., 2007.
WHAT USE IS INTELLIGENCE?
321
Figure 10.2. The distribution of RSPM scores across grades in a
representative sample of 500 Icelandic schoolchildren. From Pind,
Gunnarsdottir, & Johannesson, 2003, Figure 1, with permission from
Elsevier.
Design, Music, and Textiles, had correla¬
tions in the .4-. 5 range. This very large study
provides strong evidence for a robust rela¬
tionship between intelligence, as assessed
by a standard test, and general academic
achievement.
Correlations between battery-type tests
and academic achievement can be criticized
on the grounds that some of the subtests
in a test battery are close to a sample of
academic tasks, so what we are measuring
is the stability of academic aptitude, rather
than a more general factor of intelligence.
This interpretation cannot explain the fact
that results somewhat similar to those in
the English study have been obtained in
Iceland, using Raven’s Standard Progressive
Matrices (RSPM], which is certainly not
tied to the K-12 curriculum.20 As was the
case for the English study, this study used
a very large sample, representative of the
population of schoolchildren in Iceland. Fig¬
ure 10.2 shows the results. Scores rise over
the school-age years, agreeing with our intu¬
ition that children increase in their cogni¬
tive competence as they grow older. The
RSPM scores within class levels were corre¬
lated with grades. At the seventh-grade level
20 Pind, Gunnarsdottir, & Johanesson, 2003.
the correlations were .75 for overall mathe¬
matics grades and .64 for overall Icelandic
(language arts) grades.
The British and Icelandic studies agree
with each other well, even though the
particular tests of intelligence used were
quite different. They answer Macintosh’s
legitimate concern. Range restriction effects
do operate in small, localized studies, and
therefore corrections for range restriction
are appropriate. The correlation between
intelligence test scores and academic
achievement in the K-12 system, calculated
across large units, such as districts, is at least
.50 for the core academic courses, such as
language arts, science, and mathematics, and
somewhat lower for courses in vocational
topics and in the arts. Within smaller units,
such as a school or a class, the correlation
will drop to around .30, due to restriction of
range.
In the K-12 system cognitive test scores are
used to identify students whose low scores
indicate that they may need to be assigned to
special education classes. Tests are also used
to assign students to accelerated programs
for the gifted. In both cases other factors are
also considered. The majority of students fall
somewhere between these extremes, and for
them the test scores do not matter, for no
322
HUMAN INTELLIGENCE
Table 10.1. Correlations between tests used for college/university selection, the
ASVAB general factor or AFQT, and Raven's Advanced Progressive Matrices
Test SAT ACT ASVAB
ACT .87 [1)
ASVAB .87 (3), .92 (1] .77 [4), .90 (1)
Raven's Advanced Progressive Matrices .71 (2} .61 (3]
Source: Data sources are (1) Coyle & Pillow, 2008 (NLSY97 data]; (2] Frey & Detterman, 2004;
(3) Koenig, Frey, & Detterman, 2008.
decisions are made on the basis of these
scores. In college and university entrance
decisions test scores matter, a lot, across the
entire range of scores.
10.2.2. Intelligence in the Post-Secondary
System
Since World War II American colleges and
universities have incorporated two major
testing programs, the SAT and the Amer¬
ican College Testing Program (ACT], into
the admissions process. These tests are val¬
idated regularly, by correlating test scores
with first-year grade point average (GPAi],
cumulative grade point average (GPAC), or
probability of graduation within a specified
period of time after matriculation, usually
four to six years. The use of the tests is
not without controversies, a point that was
made earlier (section 2.7.3). We concentrate
on technical rather than policy issues here.
Recall, from the discussion in Chapter 2,
that the first portion of the current SAT,
referred to officially as the SAT-I, contains
sections stressing verbal comprehension and
logical reasoning. By tradition (and officially,
in earlier versions), these two sections have
been referred to as the SAT-V and SAT-
M. I will continue this usage. Both tests
represent attempts to evaluate comprehen¬
sion and reasoning without tying questions
to specific high school curricula.
The American College Testing program
takes a different approach. It develops tests
that are specifically tied to curricular mate¬
rial, such as history and mathematics. The
idea is to predict what a student will learn
in college by determining how much he or
she has learned in high school. The second
part of the SAT program, the SAT-II, does
the same thing.
Although there have been arguments
about which approach is better,21 the tests
could be interchanged in an academic
selection program without changing accep¬
tance and rejection decisions very much.
Table 10.1 presents estimates of the cor¬
relations between the SAT, ACT (sum¬
mary score), and the general factor derived
from the ASVAB, which is closely approx¬
imated by the Armed Forces Qualify¬
ing Test (AFQT). The correlations with
Raven's Advanced Progressive Matrices are
also included, in order to show the rela¬
tion between the educational tests and an
avowed marker for general intelligence, g.
The correlations are quite high. The cor¬
relation between the SAT and the ACT
approaches the reliabilities of the two
tests. This suggests that the true correla¬
tion between the two tests is one! A study
using NLSY97 data found that both aca¬
demic aptitude tests had loadings of about
.9 on a general factor derived from the
ASVAB.22 The finding is important because
the ASVAB general factor is a measure of
crystallized intelligence (Gc), rather than
of Gf.2?
The need to distinguish between Gf and
Gc in college students is shown by the fact
that the correlations between matrix tests
and the SAT and the ACT are in the .6-
.7 range.24 This is about what one would
21 Lemann (1999) discusses the dispute in some detail.
It has been carried forward to this day.
22 Coyle & Pillow, 2008.
23 Roberts et al., 2000.
24 Frey & Detterman, 2004; Koenig, Fry, & Detterman,
2008.
WHAT USE IS INTELLIGENCE?
expect, because the fact that Gc and Gf
are themselves correlated in the .5-7 range,
depending upon the sample. Because aspir¬
ing and attending college students represent
roughly the upper two-thirds of the general
population, in terms of cognitive skills, one
expects the general factor to be somewhat
weaker among this group than among the
population at large. (See Chapter 3 for elab¬
oration.}
Do the tests work? Several appropriately
designed large studies of the SAT have pro¬
duced consistent results. The correlation
between SAT scores and GPAi is approxi¬
mately .35 in students who have been admit¬
ted, and who therefore have both SAT
and GPAi’s available.25 This is the uncor¬
rected correlation in the selected popula¬
tion, whereas what is needed is predictive
correlation in the applicant population. An
extensive study by Paul Sackett and his col¬
leagues at the University of Minnesota, in
which they conducted a meta-analysis of
previous studies, shows quite clearly what
the situation is.
Sackett and his colleagues analyzed data
provided by the College Board for 41 colleges
and universities where the SAT was used
in 1995-97. Over 155,000 test takers were
involved. The researchers calculated three
SAT-grade correlations. They were:
1. rs - the correlation between SAT and
GPAi in admitted students, calculated
within institutions and then averaged.
rs = -35-
2. rpi - rs corrected for restriction of range
within the applicant population for each
institution, and then averaged. This is
the predictive correlation that would
be of interest to admission officers.
rpi = -47-
3. rp2 — rs corrected for restriction of range
of SAT scores across all institutions.
This can be thought of as the predic¬
tive correlation to be used to determine
the benefit of using the test across all
participating institutions. rpz = .53.
25 Geiser & Studley, 2002; Kobrin et al., 2008; Sackett
et al., 2009.
323
Freshman grade point averages indicate
a student's initial reaction to college. What
about predicting later performance or grad¬
uation? Beyond the first year there is great
variation in the courses college students
take, and there are also substantial differ¬
ences in grading practices across disciplines.
This muddies the situation.
There is a negative correlation between
the SATs of students within an academic
program and the mean grade point assigned
by that program. This is because mathemat¬
ics and science programs, which assign rela¬
tively low grades, tend to draw the students
with the highest SATs, while humanities
and education programs, which assign high
grades, draw students with lower SATs. The
effect is quite large. A study involving over
200,000 students from 38 public universities
during the 1990s26 found that the difference
in SAT scores between the discipline with
the highest entering scores, engineering, and
the one with the lowest scores, education,
was .92 standard deviation units.27 The neg¬
ative correlation between the rigorousness
of grading within a discipline and the SATs
of the entering students will reduce the cor¬
relation between overall GPAs and entering
SATs, calculated over the institution as a
whole.
The probability of graduation behaves
much like, but not exactly like, GPAi.
Herrnstein and Murray’s analysis of the
NLSY79 database showed that, as of the
1980s, approximately 70% of the survey par¬
ticipants in the top decile of AFQT scores
obtained bachelor’s degrees. This fell to
30% in the eighth decile, and to 10% in
the fifth decile.28 A detailed report from
the College Board29 found that graduation
26 Kroc et al., 1997.
27 This is a conservative estimate, based on the
assumption that the within-discipline standard devi¬
ation is equal to the population standard devia¬
tions. The assumption is very unlikely to be valid.
The effect of interdisciplinary variation would be to
reduce the variance (and hence the standard devia¬
tion] within disciplines. The upshot would be that
less than 18% of the entering education students
would be expected to have SAT scores above the
engineering mean.
28 Herrnstein Murray, 1994, p. 37.
29 Kroc et al., 1997.
HUMAN INTELLIGENCE
324
rates are nonlinearly (logistic) related to
an index composed of SAT, High School
Grade Point Average (HSGPA), and sev¬
eral demographic variables, including gen¬
der and race. People with relatively low
scores on the index generally were unlikely
to graduate; people with high scores were
highly likely to graduate; and the prob¬
ability of graduation changed markedly
between “low average" and “high average"
scores.
As was the case for the K-12 system,
the findings on the correlation between
test scores and college/university success are
strikingly consistent. The SAT, the most
widely used test, has a predictive validity
of about .5. This is probably an underesti¬
mate of the correlation between the SAT
and an abstract measure of academic abil¬
ity. Because students with high SAT scores
are more likely to enroll in “tough-grading”
courses than students with low SATs the
SAT-GPAi correlations will be depressed
below what they would have been if all stu¬
dents took the same courses.
Are these correlations really enough to
justify test use? Answering that question
requires a brief discussion of the statistics
of personnel selection.
10.2.3. Cognitive Tests and Selection
Decisions
The college/university admissions process is
an example of a personnel selection decision.
How useful are entrance examinations, such
as the SAT, in making such decisions? This
raises the question of how high a correlation
has to be in order to be useful in practice,
whether or not it is “statistically significant.”
This depends upon how the correlation is to
be used.
A widely cited way of evaluating the size
of a correlation is to square it, and then
to report it as the proportion of variance
accounted for in either variable by predic¬
tions using the other variable. In the admis¬
sions case, would be the proportion of
variance in grades that could be associated
with variance in an admissions test - approx¬
imately ,52 = .25. Multiplied by 100, one
could say that 25% of the variance in grades
is accounted for by variance in the examina¬
tion. If, as is often (and erroneously) done,
this logic is applied to the uncorrected cor¬
relation between grades and test scores, in
the population of admitted students, .352 =
.11, so 11% of the variance of grades is asso¬
ciated with variance in test scores - which
does not seem high. However, that is mis¬
leading.
If the selector uses a screening exam¬
ination, it is possible to predict aptitude
(grades or workplace performance) and
accept people in order of predicted perfor¬
mance. Unless prediction is perfect (pre¬
dictive validity = 1) people with the same
predicted performance usually turn out to
have different actual performances. Stu¬
dents with identical SATs do not all have
identical grades. In statistical terms, there is
variance around the predicted performance
level, and the greater the variance, the less
accurate the prediction. However, variance
around the predicted performance can never
be greater than the variance in the appli¬
cant population. So variance in the applicant
population can be used to scale the extent
to which the prediction is not accurate. The
ratio I = (variance around predicted value
of aptitude) /(variance of aptitude in appli¬
cant population) represents an “inaccuracy”
index, relative to the inaccuracy that would
be achieved without using a selection exam¬
ination. It follows that the complement of
I, 1 — /, is an index of accuracy. It can
be interpreted as the relative reduction in
inaccuracy achieved by using a predictor
test. The I index is related to predictive
validity, as defined in section 10.1, by the
equation
Tp ~ 1 1 (10.5)
where p = 1 or 2, depending upon whether
you are interested in within -institution or
across-institution predictivity. Multiplied by
100, r2 is the percent increase in efficiency
achieved by using a screening examination.
If, as is the case, rp = .5, the increase in
efficiency is 25%.
WHAT USE IS INTELLIGENCE?
}25
Figure 10.3. Test accuracy and rejection rate interact to produce
quality acceptances. The expected value of aptitude for an accepted
candidate (student or worker), measured in terms of the percentile
of aptitude in the applicant population and shown as a function of
the rejection rate and the predictive validity of a screening
examination.
At this point we can see an argument
brewing between the admissions commit¬
tee and the rejected applicants. Suppose an
applicant is rejected, and then learns that
among accepted applicants (students) the
correlation between SAT and grades is only
.35. How dare the committee reject an appli¬
cant on the basis of a test that is only 11%
better than chance?
The committee’s first reply can be that
the correlation is not really .35; it is .5. The
applicant’s rejoinder is that a 25% improve¬
ment over chance still is not good enough.
But this is not the admissions committee’s
real argument.
The admissions committee is not inter¬
ested in the accuracy of individual predic¬
tions; it is interested in selecting the best
possible entering class. Suppose that the
institution has room for only 10% of its appli¬
cants (a rejection rate of 90%). Insofar as is
possible, the committee wants to select the
top 10% of the applicants in terms of aca¬
demic aptitude. However, the committee
knows only the top 10% of the test scorers.
If rp = 1, the two "top ten percents" will be
the same people; to the extent that rp is less
than 1, there will be some disagreement.
The success of the selection process will
be determined by both the accuracy of the
test, rp, and the rejection rate. If the rejection
rate is zero, everyone who wants to enter
gets to enter. The accuracy of the test does
not matter, because no decision is going to
be made using the test score. At the other
extreme, suppose there is just room for one
person. The person accepted will be the one
with the highest test score, and the proba¬
bility of that person being the person with
the highest aptitude in the applicant pool
will depend upon the accuracy of the test.
Between these two extremes, the ex¬
pected quality of accepted applicants is
determined by an interaction between rp and
the rejection rate. The form of this interac¬
tion is shown in Figure 10.3, using Sackett's
two estimates of the predictive correlation
as examples. If the rejection rate is low, the
value of the predictive correlation matters
very little. If the rejection rate is high, it mat¬
ters a lot. For example, if the rejection rate is
90%, as it is for some of our elite universities,
the use of an entrance examination with a
predictive correlation of .47 can improve the
mean level of aptitude in the entering class
from the fiftieth percentile of the aptitude
in the applicant population (no test used) to
about the seventy-seventh percentile.
Exactly the same reasoning applies to
industrial hiring. If the rejection rate is high,
a screening examination with predictive
validity in the .4-. 5 range can substantially
326
HUMAN INTELLIGENCE
o^~ c> /P
Figure 10.4. Median weekly earnings in 2008 as a function of level of
education. Source : US Bureau of Labor Statistics. HS - high school
diploma.
improve the selection process, as seen by the
employer.
Note that quality has been defined in
terms of the quality available in the appli¬
cant population. Personnel selection has to
operate with this constraint; you cannot
select people who do not apply. Any recruit¬
ment technique that improves or diminishes
the distribution of aptitudes in the appli¬
cant population will affect the quality of
the selected applicants - either the student
body or the workforce. What this effect will
be will depend upon the amount of change
in applicant aptitude and upon the effect
of added or reduced recruitment upon the
rejection rate.
10.2.4. Alternatives and Augmentations
to the Use of Test Scores in College
Entrance Decisions
Chapter 9, section 5, contained a century-
old quotation from Theodore Roosevelt
about the economic value of education.
Figure 10.4 shows income figures, as of 2008,
as a function of level of education com¬
pleted. Roosevelt’s remark rings true. How
we decide who gets to go to college makes
a tremendous difference in who has eco¬
nomic and social opportunity. Therefore,
it is understandable that college entrance
examinations such as the SAT and the ACT
have received considerable scrutiny.
SAT scores are positively correlated with
parental SES. This has led some to fear that
using the SAT simply identifies applicants
who have the social and financial resources
to complete the undergraduate program.
Giving these applicants preference in college
admission will therefore exacerbate inher¬
ited social advantages, something that is
generally not considered a good thing in
a democracy. (Paradoxically, the SAT was
originally designed to reduce these advan¬
tages! See the discussion of the SAT in
Chapter 2.)
To what extent is this concern war¬
ranted? The way to investigate the ques¬
tion is to examine the partial correlation
between grades and test scores, equating
for SES. If the SAT is a proxy for SES,
the partial correlation should approach zero.
However, it does not. An analysis of the
forty-one-institution data collected by the
College Board found that the partial corre¬
lation, based on an analysis of over 155,000
students, is .44 - very little different from
the predictive correlation without consider¬
ing SES, .47. The analysis can be reversed,
to see if SES is associated with first-year
grades, after equating for test scores. When
this is done the correlation between SES and
WHAT USE IS INTELLIGENCE?
327
GPAi drops from .31 to .05. Similar results
were found in the meta-analysis of previous
studies.30
These values are consistent with the
assumption that parental SES does influence
undergraduate performance, but that it does
so as a distal variable. SES exerts its influ¬
ence by influencing traits that are impor¬
tant for success as an undergraduate and that
are measured by the SAT. Presumably these
traits are what we mean by intelligence.
High school grade point average
(HSGPA) has also been used as a predictor
of GPAi and college graduation. A study
conducted in the University of California
system found that, averaged over the incom¬
ing freshman classes of 1996-99, the SAT-I
could predict 13% of the variance in GPAi
[r = .36), HSGPA could predict 15.4% (r =
.39), and the two of them together could
predict 20.8% of the variance (R = .46]. 31
These correlations, which are consistent
with the data from the forty-one-institution
study, have not been corrected for range
restriction. As far as accuracy of prediction
is concerned, the appropriate thing to do is
to combine the entrance examination and
HSGPA into a single index.
As is true of all cognitive tests, the SAT
has been designed to measure “can do”
aspects of cognition. Cumulative indices of
performance, such as HSGPA and GPAi,
also tap “will do” aspects of performance,
such as study habits and perseverance. We
have seen this already for HSGPA; the
same thing is true for GPAi. 32 The fact that
HSGPA and the SAT, combined, do bet¬
ter than either alone is further support for
an expanded definition of intelligence, to
include skill in allocation of effort over the
long haul, outside of the conventional test¬
ing paradigm.
We have been concerned here solely with
the relation between intelligence and cog¬
nitive performance after matriculation. We
need to remember that this is a limited
view of only one aspect of the admissions
30 Sackett et al., 2009.
31 Geiser & Studley, 2002.
32 Crede & Kuncel, 2008.
decision. Admissions policy makers also con¬
sider other ways in which prospective stu¬
dents may contribute to the university.
These vary from maintaining family ties
to an institution (the “legacies” that pro¬
duce endowments on which some univer¬
sities rely) to a student’s athletic abilities.
Policy makers are also influenced by a desire
to balance male-female ratios and to pro¬
mote racial and ethnic diversity in the stu¬
dent body. Considering the appropriate role
of these goals in student admissions would
take us far beyond a discussion of the value
of intelligence in education.
10.2.5. Post-Graduate Education
In 1837 a twenty-three-year- old named
Edward Cree received a license to practice as
an apothecary from the University of Edin¬
burgh. Having a desire to go to sea, he was
appointed a Surgeon in the Royal Navy.
Ten years later he took some time off from
the Navy to complete his M.D.33 The same
sort of thing happened in the United States.
The Greenfield Village “living museum,” part
of the Henry Ford Museum, contains an
account of a mid-nineteenth-century physi¬
cian who “studied medicine awhile" at the
University of Michigan and Case- Western
Reserve University, decided he had learned
all about medicine that he needed to know,
and set up his practice on the northwestern
frontier.
Today we are a bit more formal. We
demand completion of programs for entry
into many professions. No credit is given
for attendance. The rewards for completion
can be considerable. Just as there is a 50%
increase in income for going from the High
School degree to the Bachelor's, there is
another 50% by going from the Bachelor’s
to the Doctorate (Figure 10.4).
A variety of cognitive tests are used as
screening examinations for post-graduate
educational programs. It is difficult to say
anything comprehensive about their valid¬
ity, for the importance of grades in gradu¬
ate education varies tremendously with the
33 Cree, 1982.
328
HUMAN INTELLIGENCE
Table 10.2. Odds ratios comparing probability of graduation for the top and bottom
halves of admission test scores in an entering population of post-graduate students
Test
Typical Manner of Use
Odds Ratio
Graduate Record Examination
Entrance into Ph.D. programs in
many fields of study
2.3:1
Miller Analogies Test
Entrance into Ph.D. programs in
many fields of study
2.2:1
Law School Admissions Test
Entrance into Law School
1.4:1
Graduate Management Admissions Test
Entrance into MBA programs in
business and management
1.6:1
Medical College Admissions Test
Entrance into Medical School
1.7:1
Source: Data excerpted from the supporting online material for Kuncel & Hezlett, 2007, Table si.
Reprinted with permission from AAAS.
program. It is my impression [but no more
than that] that grades are taken fairly seri¬
ously in professional programs such as Law,
Business, and Medicine, and are regarded as
incidental to research participation in most
science programs. A validity measure that
does not require equating grades across pro¬
grams is the accuracy with which high scores
predict program completion. This is mea¬
sured by the odds ratio for program comple¬
tion, which is defined as
Completion rate for
students whose entrance
, „ scores are in the top half
Odds Ratio = - - - - - .
Completion rate for
students whose entrance
scores are in the bottom
half
Table 10.2 shows the odds ratios for a vari¬
ety of entrance examinations used as part
of the screening examinations for entry into
various graduate schools. The odds ratios
vary from a low of 1.4 (for the Law School
examination] to a high of 2.3 (Graduate
Record Examination]. Having a test score in
the top half of applicants is associated with
at least a 40% improvement in probability of
graduation, compared to test scorers in the
bottom half.
Although having a post-graduate degree
clearly pays off, completing a post-graduate
course often entails considerable financial
and personal sacrifice in the short term. The
information in Table 10.2 is of as much use
to an accepted applicant, trying to decide
whether to enter graduate school, as it is to
an admissions officer.
10.3. The Workplace
Do tests of intelligence predict performance
in the workplace? Here is a claim by three
industrial-organizational psychologists .
Many lay people, as well as social scien¬
tists, subscribe to the belief that the abili¬
ties required for success in the real world
differ substantially from what is needed to
achieve success in the classroom. Yet, this
belief is not empirically or theoretically sup¬
ported. A century of scientific research has
shown that general cognitive ability, or g,
predicts a broad spectrum of important life
outcomes, behaviors, and performances.
Kuncel, Hezlett &£ Ones, 200 4, p. 148
Putting Kuncel and colleagues' proposition
more argumentatively, this is a case where
the public (and many social scientists] have
made up their mind, so please do not con¬
fuse them with facts.
WHAT USE IS INTELLIGENCE?
329
Linda Gottfredson is ready to plunge
ahead, whether she is believed or not:
In no realm of life g is all that matters , but
neither does it seem irrelevant in any. In
the vast toolkit of human abilities, none has
been found as broadly useful - as general -
as g.
Gottfredson, 2002, p. 332
Gottfredson is right, but Kuncel and col¬
leagues are right to be concerned that the
facts are a hard sell. Some of the reasons why
are captured in a third quote, this time by
J. Raven, the son of the J. C. Raven who
developed progressive matrix testing and
himself a prolific researcher on intelligence.
Given his pedigree, one might expect J.
Raven to take Gottfredson’s position, but
he is rather hesitant.
In the workplace and in the educational
system numerous other qualities are impor¬
tant but remain invisible if one utilizes only
tools developed within the traditional mea¬
surement paradigm, focuses mainly on con¬
ventional criteria of job performance, and
accepts assumptions about the functional¬
ity of hierarchical organization of work¬
places and society.
J. Raven, 2008c, p. 43 2
J. Raven further argues that the impor¬
tant things determining job performance
are not general cognitive power, but rather
the specific skills and the motivation that a
person brings to work. He also points out
that evaluations of both job performance
and academic success take place in con¬
strained situations. The constraints of the
situation may be just as important as cog¬
nitive capabilities in determining behavior.
Constraints on job performance vary widely
across the workplace, while academic con¬
straints are more uniform. This argument is
worth developing.
In the academic setting there is a reason¬
ably clear-cut criterion for success - how
well does a student know the material stated
in the curriculum? When implicit objectives
like “teaching a student how to think” are
introduced, agreement over the criteria for
success vanishes.
10.3.1. Some Evidence from Studies
of Military Enlisted Performance
In the 1980s the United States Department
of Defense conducted extensive studies of
the prediction and assessment of the job
performance of enlisted personnel.34 The
predictive measurements taken included
cognitive and personality tests and bio¬
graphical statements of interests. Occupa¬
tional assessments were similarly varied.
They included examination of service record
books (which contain job performance rat¬
ings) and records of promotions, commen¬
dations, and disciplinary actions. In addi¬
tion, both pencil-and-paper and hands-on
performance tests were given. Examinees
had to demonstrate their general skills and
knowledge as soldiers, sailors, marines, or
airmen and their proficiency in their specific
occupations. The occupations chosen var¬
ied from strictly military positions, such as
infantrymen and artillerymen, to jobs with
exact counterparts in the civilian world,
such as automobile mechanics, clerks, and
cooks.
Five dimensions of job performance were
identified. Two, general military proficiency
and technical proficiency in one’s specialty,
were “can do” measures. They evaluated
how well a person could do his or her job,
when they knew that they were being eval¬
uated. The next three factors were “will do”
measures. Discipline referred to whether or
not the individual followed regulations and
could be relied upon to be ready to do his
or her job. Leadership referred to the ability
to encourage others and to take initiative.
Fitness referred to personal bearing, appear¬
ance, and physical fitness. With the possible
exception of fitness these dimensions apply
to both the military and civilian workplaces.
Figure 10.5 shows the relation between
the five factors and measures of personality,
biographical interests, and cognitive perfor¬
mance (including scores derived from the
ASVAB). The cognitive measures were the
best predictors, by far, of the two “can do”
34 Campbell & Knapp, 2001.
HUMAN INTELLIGENCE
330
Technical proficiency
General soldiering
Leadership
Discipline
Fitness
0 0.2 0.4 0.6 0.8
Predictive Validity
□ Vocational interest
□ Temperament/personality
■ Cognitive ability
Figure 10.5. Correlations between predictors and criterion
measures in the U.S. Army study of enlisted performance. Data
from McHenry et al., 1990, Table 4.
factors. Interest and personality measures
were the best predictors of the “will do”
aspects of job performance.
Steven Hunt, an industrial and organi¬
zational psychologist, has pointed out that
the first two steps in developing an assess¬
ment program in industry are to define the
job that you expect employees to do and
to determine how you are going to decide
whether their performance measures up to
these expectations.35 It is not reasonable to
expect anyone to excel in all aspects of per¬
formance. To the extent that the required
job skills are themselves not correlated, it
is impossible for one predictor to predict
them all. The results shown in Figure 10.5
illustrate Steven Hunt's point. “Can do” is
useless without “will do.”
Job performance is highly dependent
upon experience, because the more one
practices something, the better one becomes
at it. Expertise in complex tasks can take
years to acquire.36 Expertise implies the abil¬
ity to learn from experience.
Further military studies showed that job
performance was a joint function of expe¬
rience on the job and intelligence. Soldiers
at all intelligence levels took about eighteen
months to approach their top levels of per¬
formance, with much slower improvement
in the next two years. Soldiers with the high¬
est cognitive scores (AFQT Level I and II)
35 S. Hunt, 2007, Chapter 5.
36 Ericsson, 2003.
performed better after six months on the job
than soldiers with lower scores after forty-
two months.37
The military provides a highly structured
workplace, and the workforce is younger
than the civilian workforce. What are the
relationships between intelligence and per¬
formance in the civilian workplace?
10.3.2. Evidence from the Civilian
Workplace
Literally hundreds of studies have been
done of the relation between test scores
and job performance in the civilian sec¬
tor, using tests ranging from the ASVAB,
which takes several hours, to the Won-
derlic and the Raven tests. Two American
industrial-organizational psychologists, John
Hunter and Frank Schmidt, have conducted
a number of widely cited meta-analyses of
these results. Figure 10.6, taken from one
of their best-known studies,38 shows that in
the blue-collar, clerical, and administrative
occupations the predictive validity of gen¬
eral intelligence, averaged over all studies, is
.51 (corrected for range restriction and unre¬
liability in the job performance criterion).
The validity coefficient can be increased
by combining a measure of general mental
ability with various other assessment meth¬
ods. Predictive ability can be raised to its
37 Wigdor 81 Green, 1991.
38 Schmidt & Hunter, 1998.
WHAT USE IS INTELLIGENCE?
331
Job tryout
Job knowledge
Reference check
Peer rating
Unstructured interview
Conscientiousness test
Integrity test
Work sample
General intelligence
m Multiple R
i Correlation
Correlation corrected for range restriction and criterion
unreliability
Figure 10.6. The correlations between measures of job
performance, measures of general intelligence, and a variety
of other assessment measures. The values shown are for the
correlation between job performance and the assessment
(Correlation) and the correlation between job performance and
an optimum weighting of the assessment and the assessment of
general intelligence (Multiple R). Data from Schmidt & Hunter,
1998, Table 1.
maximum validity, .65, by combining a mea¬
sure of general intelligence with a test of
integrity. (Conscientiousness is a close sec¬
ond, at .60.) This illustrates the combined
importance of “can do" and “will do” traits.
The correspondence between the mili¬
tary and civilian data shows that the findings
are robust over different situations and dif¬
ferent methods of evaluation. The military
data was gathered by direct observation of
young adults; the civilian figures were based
on a meta-analysis of dozens of small studies,
covering all age ranges, but none as compre¬
hensive or rigorous as the military studies.
The fact is clear. General intelligence has
a predictive validity of about .50 in the
workplace, just as it does in academia. No
other method of assessment does any better.
Nevertheless, people persist in using other
techniques for predicting workplace perfor¬
mance. An examination of these alternatives
is in order.
The only type of test with predictive
ability greater than a test of general intel¬
ligence is a work sample (correlation of
.54 compared to .51). This can be used in
some situations. For instance, when musi¬
cians audition for places in major symphony
orchestras they are often asked to play their
instruments behind a curtain, so that the
judges do not know who the candidate is.
Work samples have two highly desirable
qualities: they are statistically valid, and they
are easily justified when assessment meth¬
ods are challenged. Their drawbacks are that
they can be rather expensive and that they
can be used only if the candidates for a job
have already been trained to do the job.
Combining a work sample and a gen¬
eral intelligence measure increases predic¬
tive validity to .63. The increase is not
surprising, for by combining a general intel¬
ligence measure with a work sample the
employer is simultaneously informed about
the prospective employee's general reason¬
ing powers and specific job knowledge.
In personnel selection situations a test
this accurate, combined with a high rejec¬
tion rate, can greatly increase the quality of
the employed workforce. Recall that if no
screening test is used, the average person
hired should have an ability level equal to
the fiftieth percentile (median) of the appli¬
cant population, regardless of the rejection
rate. If a predictive validity of .63 is com¬
bined with a rejection rate of 50% (half
the applicants are hired), the average ability
level of a person hired will be at the sixty-
ninth percentile of the applicant population.
An unstructured interview is an interview
in which the recruiter and the candidate
“just chat,” so that the recruiter can get a
HUMAN INTELLIGENCE
332
feel for the candidate. This is probably the
most widely used selection procedure. The
unstructured interview is not very good on
its own (r = .38, corrected) and adds very
little to the information gained in a test of
general intelligence.
A structured interview (not shown in
the figure, but included in Schmidt and
Hunter’s analyses) is an interview in which
the recruiter has decided, beforehand, what
topics are to be discussed in the interview,
and what information must be provided.
The technique requires a careful analysis
of the requirements of the position to be
filled, before searching for candidates. Struc¬
tured interviews have good predictive valid¬
ity, both on their own and when combined
with a test of general intelligence (r = .51,
R = .63).
Job knowledge is usually assessed by per¬
formance on a written test, where the ques¬
tions are chosen to reflect what a job holder
should know. This is a face-valid measure;
we can reasonably expect bus drivers to
know the rules of the road, and firefight¬
ers to know how to use various pieces of
equipment. Job knowledge is not quite as
good a predictor as is general intelligence,
but it does add to predictive validity beyond
that provided by a general intelligence score
(r = .48, R = .58). In terms of the Gf-Gc
model of intelligence, what a job knowledge
test does is assess what the applicant knows
about the particular situation in which he
or she will be working. The same idea is
captured by Robert Sternberg’s emphasis on
practical intelligence, which would include
job knowledge. Sternberg and his col¬
leagues have provided such tests, and they
have on occasion shown some incremental
validity.39
The practical intelligence tests Sternberg
and his colleagues have described are very
close to job knowledge tests. For instance,
one practical intelligence test, designed for
Alaskan hunters, asked what different pieces
of evidence mean as indicators of coming
39 See Sternberg, 2003, for a general discussion, and
Sternberg et al., 2000, for a compendium of many
of the studies.
weather.40 Such questions measure crystal¬
lized intelligence (Gc) within a specialized
context.41
10.3.3. Upper-Level Managerial and
Professional Positions
The data presented so far is based largely
on data from studies of blue-collar and
white-collar jobs, up to the lower manage¬
rial level. In this population of occupations
the correlation between general intelligence
test scores and job performance generally
rises with increasing job complexity.42 Given
this fact, it would be reasonable to expect
the correlation to be still higher for high-
level managerial, executive, and professional
positions. However, there are reasons not
to assume a straightforward extrapolation
of the results to the managerial/professional
class.
Many studies of high-level occupations
report the observed correlation between test
scores and measures of job performance,
but cannot correct for selection restriction
because there is no data on the applicant
population. This is serious, because the
selection effects are likely to be large. High-
level positions are quite competitive, and
are virtually always filled by people in the
upper quartile of the intelligence range, IQ
110 and above. It is also difficult to find a
measure of how well a professional or exec¬
utive is doing, beyond gross judgments of
satisfactory or unsatisfactory performance.
As Fortune magazine repeatedly shows in its
annual survey of executive salaries, the cor¬
relations between executive compensation
and objective measures of company perfor¬
mance are close to zero. Physicians, attor¬
neys, and other professionals are evaluated
periodically, but the ratings are often lim¬
ited to certification of competence without
any further differentiation.
It is also often hard to acquire the
required data on intelligence. People who
40 Grigorenko et al., 2004.
41 See Gottfredson, 2003a, and Hunt, 2008, for expan¬
sions on this point.
42 Gottfredson 1997, 2002.
WHAT USE IS INTELLIGENCE?
333
occupy high-level positions are busy, and
usually see no need to have their cognitive
skills evaluated. As a substitute for direct
observation, many studies look at profes¬
sional training rather than on-the-job per¬
formance. As was shown in the section on
post-graduate education, cognitive tests do
surprisingly well in predicting completion of
professional and managerial training.
There is also the problem of the multidi¬
mensionality of the criterion. People who
occupy high-level positions are typically
asked to do a number of different tasks.
These range from high-level planning to
public relations, face-to-face leadership, and
negotiations. The relative importance of dif¬
ferent tasks varies greatly across occupations
and even from time to time within an occu¬
pation. It is not surprising that there has
been a good deal of resistance to the idea
that any unidimensional measure could pre¬
dict performance at high levels. This has led
to three different approaches.
In order to evaluate highly intelligent
people using the conventional psychomet¬
ric paradigm we have to have harder tests.
Examples are the advanced version of the
Raven tests, the Raven Advanced Progres¬
sive Matrices (RAMP), and the Miller Anal¬
ogy Test (MAT), which contains difficult
verbal analogy problems. These tests do pre¬
dict job performance, with observed corre¬
lations in the .15-. 30 range, depending upon
the criterion used to evaluate performance,
and with a predictive ability on the order of
.40.45 This is somewhat below the level of
prediction obtained for higher-level skilled
work, but still a reasonable figure.
It is somewhat more enlightening to look
at a single large prospective study. During
the 1960s the Bell Telephone System, at the
time a near-monopoly covering telephone
services in the United States, used the assess¬
ment center technique to select beginning
managers. Management trainees spent sev¬
eral days in an assessment center, where they
were rated for their ability to solve com¬
plicated problems both individually and in
43 Kuncel, Hezlett, & Ones, 2004; Raven, 2008b.
groups, and also given a cognitive test sim¬
ilar to the SAT reasoning tests, along with
several personality tests. The results of these
assessments were carefully shielded from
their superiors in the company, so that the
test scores could not influence supervisors’
judgments or hiring decisions. (By contrast,
in the US military a service person’s scores
on entry tests are part of his or her service
record, and hence are available to comman¬
ders and promotion boards.) Twenty years
later the assessment center results were val¬
idated by determining whether they pre¬
dicted the level of management the can¬
didate had achieved. The cognitive test
(r = .38) was by far the best predictor.44
As would be suggested by Schmidt and
Hunter’s analyses, personality tests had
lower validity than cognitive tests, but did
add substantially to predictivity.
Because high-level performance is said
to be so multidimensional, some interest¬
ing alternatives to the conventional testing
methods have been developed. One of the
more popular of these is the situational judg¬
ment test. In this test an examinee is asked
what he or she would do in a realistic, dif¬
ficult situation. An example I particularly
like, and that has appeared in a number
of guises, is asking an applicant for a mid¬
dle management position how they would
inform their own supervisor that the super¬
visor’s pet project was not working. As the
example illustrates, an attempt is made to
design situational judgment tests that draw
on both cognitive skill narrowly defined and
the examinee’s social skills. Situational judg¬
ment tests add an additional .06 to the pre¬
dictive validity that can be achieved by a
cognitive test alone - not a large amount,
but enough to be worthwhile in a large scale
assessment program.45 It is worth noting,
though, that a situational judgment test asks
the examinee what he or she would do in a
hypothetical situation. It does not immedi¬
ately follow that that is what the examinee
would do, if placed in an actual, possibly
emotional situation.
44 Howard & Bray, 1988.
45 McDaniel et al., 2001.
334
HUMAN INTELLIGENCE
To summarize, general cognitive ability
is the best single predictor of executive/
professional-level performance, just as it
is of performance in the middle to high-
end range of the general workforce. Predic¬
tion of executive/professional performance
is somewhat less accurate than prediction of
general workplace performance. There are
several reasons why this might be so. They
include difficulties in defining and obtaining
measures of job performance, the extreme
restriction in range of intelligence among
applicants for high-level positions, and, pos¬
sibly, the fact that general cognitive abil¬
ity is a less dominating factor, compared
to other dimensions of intelligence, in the
upper ranges of cognitive competence than
in the lower (see Chapter 4). Nevertheless, it
is reassuring to know that among the movers
and shakers in our society intelligence does
count.
10.3.4. The Rewards for Cognitive Skills
in the Workplace
The previous sections have shown that there
is a positive relation between intelligence
and workplace performance, within both
military and civilian occupations. This is the
sort of information employers want to have.
From the viewpoint of an individual enter¬
ing the workforce, the question is rather dif¬
ferent. The individual wants to know what
sort of economic niche he or she is likely
to occupy, given a certain level of intelli¬
gence. As an ancillary question, what sort of
rewards can the intelligent person look for¬
ward to, in terms of either money or occu¬
pational prestige?
Determining the statistical relationship
between rewards and prestige is straightfor¬
ward. You look at the correlation between
test scores and some index of rewards. This
can be done on an individual basis, or,
as is sometimes easier to do, researchers
can determine the typical level of intelli¬
gence of people in different occupations,
and then look at the prestige and economic
rewards offered by those occupations. How¬
ever, as always, correlation does not neces¬
sarily mean causation.
In Chapter 1 I introduced the challenge
hypothesis, the idea that within genetically
prescribed limits people will increase their
intelligence in response to a cognitively chal¬
lenging environment. It has been shown, for
instance, that there are qualitative differ¬
ences in the reasoning skills of psycholo¬
gists, physical scientists, and lawyers. Psy¬
chologists, who receive substantial training
in statistics, are more sensitive to arguments
based on probabilities than are people in the
other two fields.46 Were the psychologists,
physical scientists, and lawyers attracted to
their fields because the demands of the field
matched their preferred styles of reason¬
ing, or were the styles of reasoning deter¬
mined by their experiences? The best way
to answer this question is by a prospective
study, where a person’s intelligence is deter¬
mined before he or she enters the workforce,
and then related to the person’s subsequent
work history.
During World War II the United States
Army Air Force (USAAF, the predecessor
of today's Air Force, USAF) tested large
numbers of young men who had applied to
serve as aviation officers.47 Two Columbia
University psychologists, Robert Thorndike
and Elizabeth Hagen, located approximately
10,000 of the men about twelve years after
they had been tested.48 At that time most of
the men were in their early to mid- thirties.
Table 10.3 shows the mean test scores on
general reasoning, verbal, and perceptual-
motor scales of the original test, for men in
selected occupations. These estimates were
then converted to IQ ranges. The cadets who
eventually entered those occupations that
we consider more generally intellectually
challenging, or that require a considerable
amount of education, were also the cadets
who had, at age twenty-one, scored high
on the general reasoning tests. As we scan
down the general reasoning scale we begin to
encounter white-collar office jobs, and then
various blue-collar jobs that, although they
46 Amsel, Langer, & Loutzenhiser, 1991.
47 Women were not accepted for aviation cadet train¬
ing. Some women who had already qualified as avi¬
ators in the civilian sector did serve.
48 Thorndike & Hagen, 1959.
WHAT USE IS INTELLIGENCE?
335
Table 10.3. Cognitive skills assessed in USAAF aviation cadets, shown by occupation
followed after WW II. The far left-hand column shows estimated general intelligence
measures using the conventional IQ scale. The estimate is based on a conversion of the
general reasoning score, assuming that o on that scale corresponds to 105 on the IQ scale.
Right-hand columns show mean scores achieved at approximately age twenty-one on three
different composites of a testing battery. All three scales have a mean of o and a standard
deviation of 100 in the sample of cadets.
Estimated IQ Range
Occupation
General Reasoning
Numerical
Visual-Perceptual
>115
Chemical engineer
106
42
3°
Mechanical engineer
93
34
44
Physical scientist
80
22
23
College professor
75
38
38
Civil engineer
75
31
56
Electrical engineer
65
6
9
uo< - <115
Physician
59
20
18
T reasurer/comptroller
55
96
23
Industrial engineer
44
31
34
Lawyer
39
22
-7
Personnel manager
33
18
*3
Pharmacist
29
39
“9
0
V
1
VI
ITS
O
Dentist
28
20
>5
Accountant/ auditor
28
54
-4
Optometrist
H
34
“4
Clergyman
*3
1
~l7
Airplane pilot
13
10
—1
Real estate salesman
6
17
6
ioo< - <105
Office manager
4
33
9
Insurance underwriter
3
2
“"9
Veterinarian
-8
—2
—20
Insurance claims adjuster
-13
-5
“9
Bricklayer
-24
“5
-38
Radio/TV repairman
-33
-37
21
95-<-<ioo
Hardware Salesman
-36
—12
-9
Sales clerk
-40
—22
-28
Plumber
~42
—21
“31
Carpenter
-44
-4
Police detective
-50
—26
—20
House painter
-63
—12
-24
<95
Crane operator
-66
00
1
“37
Vehicle mechanic
~72
-65
-7
Assembler (in factories)
-s.
-76
-40
Source: Data selected from Thorndike and Hagen, 1959.
may require considerable skill, are less intel¬
lectually demanding and can be learned on
the job rather than via formal education.
Some pairwise comparisons are interest¬
ing. Cadets who became physicians had
general reasoning skills similar to those of
cadets who became treasurers/comptrollers
(i.e., high-level financial managers), but the
treasurer/comptrollers had higher numeri¬
cal skills than the physicians. Cadets who
became college professors had the same gen¬
eral reasoning skills as those who became
civil engineers, but the engineers had higher
perceptual skills. This illustrates the point
336
HUMAN INTELLIGENCE
Table 10.4. Correlations between AFQT scores obtained at
age sixteen to eighteen and measures of social outcomes
when the participants were in their early thirties
Demographic
Group
Educational
Attainment
Square Root
Income
Occupational
Status Index
White men
.67
•39
•54
Black men
.48
.29
•45
White women
•59
•31
■45
Black women
•53
-44
•47
Source: Data from Scullin et al., 2000, Table 2.
that in many occupations general reasoning
skills have to be augmented by more specific
cognitive skills.
Other pairwise comparisons show that
occupations can be the same in terms of the
stress they put on different special abilities,
but differ in the level of associated general
reasoning skills. Treasurers and comptrollers
(high-level financial officers) and accoun¬
tants and auditors (financial technicians)
were both characterized by high reasoning
and numerical skills, but the cadets who
became treasurer/comptrollers had higher
levels of these skills.
While the Thorndike and Hagen study is
certainly informative, there are some aspects
that limit sweeping conclusions. The partic¬
ipants, aviation officers, were an intellectu¬
ally select group, with an estimated mean
IQ of 105. The occupations these young
officers entered, following the war, were
fairly high on the occupational scale. And
the test battery was designed for aviation
cadets. Accordingly, compared to the typi¬
cal battery-type IQ test the aviation battery
was biased toward the evaluation of visual-
perceptual skills - so much so that verbal
skills were incorporated within the general
reasoning factor. Even given these limits,
a coherent picture emerged: intelligence is
worth quite a bit.
Thorndike and Hagen studied the work¬
place of the mid twentieth century. Com¬
pared to that workplace, today's workplace
places much more emphasis on the manip¬
ulation of data and abstract representations,
rather than things.49 Have these changes in
49 Hunt, 1995; Reich, 1991; Zuboff, 1988.
the workplace changed the requirements for
cognitive skills?
To address that question we look at a
second prospective study using the NLSY79
data (see panel 9.9). Recall that in this study
a nationally representative sample of adoles¬
cents and young adults took the AFQT. In
2000 a research group at Cornell University
investigated the occupational and income
status, as of 1995-96, for over 2,400 of the
panelists who had been born in 1963-64. 50
The men and women in this study were of
approximately the same age as the former
aviation cadets studied by Thorndike and
Hagen, but in a cohort born roughly forty
years later.
In general, the Cornell group found
that there was a moderate positive cor¬
relation between AFQT score, educational
attainment, income, and occupational pres¬
tige (SEI). The correlations are shown in
Table 10.4, reported separately for four
different demographic groups, White and
Black men and women. Although the cor¬
relations are all positive, and never negli¬
gible, they do vary markedly across demo¬
graphic groups. As a rough generalization,
intelligence seems to be a more useful pre¬
dictor of future success if you are a White
man or a Black woman than if you are a
White woman or a Black man.
Table 10.4 treats educational attainment
as an outcome. Education can also be
thought of as something to be achieved
en route to further social and economic
success, rather than as an end in itself.
50 Scullin et al., 2000.
WHAT USE IS INTELLIGENCE?
337
Figure 10.7. Predicting square root income (SQRT INCOME) and
occupational prestige (SEI) in 1995 from educational attainment (in
years) and AFQT score obtained in 1980. Respondents were sixteen
to eighteen years old in 1980. Codes: WM - White men, BM - Black
men, WW - White women, BW - Black women. Data for
calculations from Scullin et al., 2000, Table 2.
Not surprisingly, educational attainment
and AFQT score are substantially correlated
in this sample, ranging from .67 (White
men) to .48 (Black men). This is consis¬
tent with the data reported in section 10.2,
relating intelligence to academic achieve¬
ment. Figure 10.7 shows that there is little
added predictive value of knowing a per¬
son’s AFQT score, once his or her educa¬
tional attainment is known.
There are three possible interpretations
of this finding. One is that education is
the proximal variable that determines socio¬
economic outcome and income, while intel¬
ligence acts as a distal variable, determin¬
ing education but then playing no further
role. The other is that intelligence acts
as a proximal variable, and that education
serves as an additional statistical marker
for intelligence beyond the AFQT score.
These explanations can be discriminated by
contrasting the partial correlations between
AFQT and the outcome variable, occupa¬
tional prestige or income, holding educa¬
tion constant, or between education and the
outcome variable, holding AFQT score con¬
stant. I calculated these and found that for
income the partial correlations for education
are generally larger than for AFQT score,
but that the differences are not great. The
results for occupational prestige are strik¬
ing. The partial correlations for education
given AFQT score are .43 for Blacks (both
men and women), and .38 and .28 for White
men and women, respectively. The corre¬
sponding values for AFQT given education
range from .16 (White men) to .20 (White
women). Evidently education and intelli¬
gence are collinear predictors of income.
Intelligence acts as a distal variable, exerting
its influence through education, which then
permits entry into prestigious occupations.
The educational effect seems to be stronger
for Blacks than for Whites.
Another way to determine what the
workplace is willing to pay for intelligence
is to examine the test scores of people who
apply for jobs in various occupations. This
carries with it the defensible assumption
that the applicants exert self-selection. Peo¬
ple without college degrees generally do not
apply for entry-level executive positions.
Gottfredson used a number of stud¬
ies of intelligence test scores in job appli¬
cants to construct a “life’s chances” chart
for various occupations.51 She associated IQ
equivalents with five classes of occupation,
ranging from what she regarded as “slow,
supervised” work to "gathers own infor¬
mation." Table 10.5 lists occupations cited
by Gottfredson, along with her estimates
of the typical IQ score for an applicant.
51 Gottfredson, 1997.
338
HUMAN INTELLIGENCE
Table 10.5. Gottfredson's examples of intelligence levels (on IQ scale) associated with
different occupations and techniques of information processing. The columns to the
right of Gottfredson's figures show the intelligence-level estimates obtained from
Thorndike and Hagen's study of the careers of USAAF aviation cadets, approximately
fifty years earlier (Table 10.4) and the range of Wonderlic scores of applicants reported
in the Wonderlic Corporation's Normative Report (April 2007) for the WPT-R, revised
in 2003. As occupations used in the norming for the WPT and WPT-R, a comparable
profession to Gottfredson's "typical profession" was used. Wonderlic scores have been
converted to IQ units using the conversion 2 * Wonderlic score + 60 (Dodrill, 1981;
Dodrill & Warner, 1988).
Training and
Qualification
Method
Typical
Position
Gottfredson’s
Estimated
Typical IQ Score
Estimates Based
on Thorndike
Hagen Data
WONDERLIC
with Comparison
Occupation
Explicit, hands-on
training
Assembler
00
0
<95
88-104 Electro¬
mechanical
assembler
Mastery learning,
hands-on training
Police officer
95-1°5
95-100
100-114 Police and
sheriff officer
College formal
instruction
Accountant
110-120
105-110
102—120
Accountant
Graduate
Attorney
120 +
110-115
110-124 Executive
instruction,
gathers own
information
The table also includes data for comparable
occupations, based on the Thorndike and
Hagen data, taken forty years earlier, and
for the 2003 revision of the WONDER¬
LIC test, the WPT-R. There is a striking
similarity between Gottfredson's estimates,
the WONDERLIC estimates, and the esti¬
mates that Thorndike and Hagen had made
forty years earlier. The data was gathered
using different sampling methods, in dif¬
ferent workplaces separated by over half
a century, and the tests used were quite
different.52 Nevertheless, the estimates are
basically the same. Comparing Tables 10.4
and 10.5, it appears that the role of intelli¬
gence in today’s workplace is very much the
52 Thorndike and Hagen’s general reasoning factor was
extracted from a battery of subtests that took several
hours to complete. Gottfredson’s estimates were
based on data from the Wonderlic Personnel Test
(WPT) The current Wonderlic estimates are based
on the revised version, WPT-R, with data from 2003.
See Chapter 2 for a description of the WPT.
same as it was before the term "information
technology” was invented.
A second retrospective study provides
a more comprehensive look at general
levels of occupational accomplishment,
but does not provide data on individual
occupations.
In the late 1980s the Center for Dis¬
ease Control conducted a follow-up study
of the health of people (primarily men) who
had served in the military during the Viet¬
nam War era (1967-76). A large number of
veterans were contacted and asked to par¬
ticipate in extensive physical and mental
testing. Helmut Nyborg, a Danish psychol¬
ogist, and Arthur Jensen related the veter¬
ans’ current occupational status to their edu¬
cational attainment and to a measure of g
extracted from their test scores, as of the
late 1980s. 53 Figure 10.8 shows the results
53 Nyborg & Jensen, 2001.
WHAT USE IS INTELLIGENCE?
339
Income
Occupational Status
Together
Black
White
0 0.1 0.2 0.3 0.4 0.5
Correlation (r) or multiple correlation
■ Together
■ Ed
■g
■ Together
■ Ed
g
Figure io.8. Correlations and multiple correlations between income
(top] and occupational status (bottom] and general intelligence and
educational level. The calculations are based on Table 3 of Nyborg &
Jensen, 2001.
of Nyborg and Jensen’s analysis. Data are
shown separately for black and white veter¬
ans, as they differed on educational, intelli¬
gence, occupational, and income measures.
In this sample the intelligence test score
was a slightly better predictor of income
than was educational achievement, while
the reverse was true for the NLSY data.
It would be unwise to make very much
of this. The samples were different, the
tests were different, and the NLSY analy¬
sis attempted to predict future accomplish¬
ment from measures taken in adolescence,
while Nyborg and Jensen related current test
performance to current accomplishment. In
any case, the similarities are far greater than
the differences.
Both educational level and intelligence
are substantial predictors of accomplish¬
ment in the workplace. The two are highly
correlated. This is hardly surprising. Both
measures are based on the development and
display of cognitive skills.
10.3.5. What the Jobs Demand
Another way of determining the value of
intelligence in the workplace is to analyze
the job requirements of a large number of
occupations, and infer what this means in
terms of the demands on intelligence. The
first step is to make an analysis of the relative
value of cognitive skills for different jobs.
The US Department of Labor (DOL)
maintains an elaborate job counseling ser¬
vice, in which it describes over 12,000
jobs and rates the extent to which they
require certain skills. The skills rated range
from general reasoning ability to finger dex¬
terity. The rating system was originally
34°
HUMAN INTELLIGENCE
120
115
•i no
105
100
V N
GATB Scale
■ Physical Science
□ Equipment Operation
■ Law
□ Child & Adult Care
Figure 10.9. Recommended GATB cut points for four occupations,
taken from high-status and lower-status patterns, within the class of
occupations dealing with physical or with social relationships.
GATB scales are G - general reasoning, V - verbal, N - numerical,
and S - spatial. Data selected from Gottfredson, 1986, Table 2. In
cases where no value is specified the DOT analysts had made the
judgment that virtually anyone would have sufficient ability to do
the job. For instance, there is no cut point for spatial reasoning
required for the law, and no cut point for general reasoning required
for heavy equipment operation.
incorporated into a descriptive volume
called the Dictionary of Occupational Titles
(DOT). The DOT has been superseded
by an on-line, interactive system called
0*NET. 0*NET is a considerable expan¬
sion over the DOT, designed primarily to
help job seekers. As a side benefit, it con¬
tains a massive amount of data available
to researchers interested in issues involving
workforce skills.
Gottfredson utilized the original DOT
system to construct a “space” of jobs.54 Her
analysis coordinated job ratings with data
on the test performance of people who had
applied for, or were occupying, a variety of
jobs. She identified five classes of occupa¬
tions - those dealing with physical relations,
social and economic relations, maintaining
bureaucratic order, and performing. (She
also had a small class of “leftover” occupa¬
tional patterns that will not be dealt with
further.) Within each of these classes she
identified the patterns of aptitudes required.
For example, within the class of occupa¬
tions dealing with physical relations there
was a cluster of occupations that dealt
54 Gottfredson, 1986.
with research and design (e.g., physicist,
engineer) and a cluster that dealt with
building, maintaining, or operating physical
objects (e.g., equipment operators, crafts¬
men). Within the class dealing with social
and economic systems one cluster dealt
with research and design (including social
research, law, and finance), while another
dealt with providing service to individuals
(including hospitality services, and child and
adult care).
Individual jobs within a cluster could
be associated with a pattern of abilities,
as defined by the DOL’s General Apti¬
tude Test Battery (GATB), which was in
use at the time. The DOL used these val¬
ues to recommend minimum values (cut
points) for a job along each of four GATB
dimensions: general reasoning, verbal rea¬
soning, numerical skills, and spatial skills.
Figure 10.9 provides examples for four occu¬
pations, two from Gottfredsons's class of
occupations dealing with physical relations
and two from the class dealing with social
relations. Within each class one occupation
was selected from Gottfredson’s high-status
cluster, and the other from a lower-status
cluster. Physical scientists and lawyers,
WHAT USE IS INTELLIGENCE?
341
250,000
-2 200,000
o
Imputed Intelligence (S.D. Units)
Figure 10.10. The ninetieth, fiftieth (median) and tenth decile of
incomes in various occupations, plotted as a function of the imputed
intelligence demands for the occupation. Data from US Census
Bureau and from Hunt and Madhyastha's analysis of job demands.
high-status occupations from the physi¬
cal relations and social relations clusters,
have similar high-level patterns except that
lawyers are not required to have spatial rea¬
soning skills, and do not have quite as high
scores on other scales as the physical scien¬
tists do. Equipment operators have to have
a certain minimal level of spatial reasoning
skills, while personal caretakers have to have
a minimal level of general reasoning skills.
Gottfredson found that general intelli¬
gence was by far the biggest driver of varia¬
tions in cognitive skills across occupations.
Verbal, numerical, and spatial skills were
important in some occupations, but they
accounted for much less of the variation
in the descriptions of occupational require¬
ments than did differences in requirements
for general reasoning.
Tara Madhyastha and I have analyzed
the modern 0*NET data (as of 2008)
using different techniques than Gottfred¬
son had used, but intended to answer basi¬
cally the same questions. In general, our
analysis of cognitive demands agreed well
with Gottfredson's.55 We found that the rat¬
ings of skills could be described by a two-
dimensional space, where one dimension
was a general reasoning factor and the other
was a bipolar factor, indicating whether the
job emphasized verbal or perceptual-motor
55 Unfortunately it is not possible to compare our
analysis to the data obtained by Gottfredson or by
Thorndike and Hagen, because our analysis was at
a much finer level of detail.
skills. We also found a smaller factor indi¬
cating the extent to which a job required
numerical skills. The factors were not statis¬
tically independent, because most jobs that
require a high degree of general reasoning
also require fairly high verbal skills, just as
Gottfredson had noticed. This is not surpris¬
ing; people who hold intellectually demand¬
ing jobs usually have to communicate with
other people. In spite of comic strip stereo¬
types, real computer programmers spend a
great deal of time describing what their pro¬
grams do, and how they fit into suites of
programs developed by other people.
Our analysis also allowed us to relate
the incomes associated with different occu¬
pations to the cognitive demands of those
occupations. The relations to general intelli¬
gence are shown in Figure 10.10. The income
associated with occupations increased as the
occupations demanded higher levels of intel¬
ligence. However, there appeared to be a
nonlinear increase; incomes are fairly flat
over occupations that require slightly more
than average intelligence. Incomes are more
closely related to demands on intelligence if
the occupation requires an IQ of more than
about 105 (g > .3 on the standard devi¬
ation scale). What is even more striking
is the range of incomes associated with a
given level of cognitive demand. This is
clearly shown in the figure. The range in
annual income between the 10% best-paid
and 10% worst-paid occupations with a cog¬
nitive demand of 100 (o in standard deviation
HUMAN INTELLIGENCE
342
units] was about $65,000 (2005 dollars). The
range for jobs with a cognitive demand of 116
[1 in standard deviation units) was roughly
$140,000.
Figure 10.10 also shows that the distribu¬
tion of income is markedly positively skewed
at all levels of intelligence. The absolute dif¬
ference between the median and the nineti¬
eth percentile is always greater than the
absolute difference between the median and
the tenth percentile. The extent of the skew
increases with increasing imputed intelli¬
gence, a trend that begins much sooner than
the rise in median income as a function of
imputed intelligence.
10.3.6. Summary: The Role of Intelligence
in the Workplace
The quotations that began this section,
from Gottfredson and from Kuncel and col¬
leagues, were accurate. Intelligence predicts
a person’s job status and income better than
any other trait that has been studied. This
leaves us with two questions: why does the
association exist, and why is it that the asso¬
ciation is so widely denied by people who
have not studied the topic?
The influence of intelligence is undoubt¬
edly mediated in part by education. This
is especially true across fields, for the vast
majority of the more lucrative occupa¬
tions have, as an entry requirement, at
least a college education. The best-paid
professional occupations require substan¬
tial graduate-level training. It is increasingly
the case that the entry routes to skilled
trades, such as auto mechanic, involve aca¬
demic certification through community col¬
lege or other professional training programs.
Because educational attainment is strongly
related to intelligence, any variable that is
correlated with educational attainment will
also correlate with intelligence.
In the case of the professions education is
essential; no one wants to have a surgeon (or
an airline pilot) who is learning basic skills
on the job. To the extent that intelligence
is required to get through a rigorous train¬
ing program; intelligence and education are
entwined.
Charles Murray has argued that in many
cases education is not really needed for suc¬
cess in a field, but that education is used as
a (socioeconomic?) screening device.56 To
the extent that this is true a spurious rela¬
tion between intelligence and economic suc¬
cess could be created, via a real relationship
between intelligence and educational attain¬
ment and a spurious one between educa¬
tional attainment and economic success, cal¬
culated across occupations. However, this is
not completely the case. Both intelligence
and education are important, because there
are nonzero partial correlations between
indices of workplace success and either intel¬
ligence or education, after the other has
been held constant.
The fact that intelligence is correlated
with on-the-job performance in a wide range
of military and civilian occupations, includ¬
ing ones that do not have high educa¬
tional requirements, provides further evi¬
dence that intelligence is important in itself,
not just as a facilitator of education.
10.4. The Social and Economic
Prospects at the Ends of the Bell Curve
If intelligence is an important trait in our
society, then the lifetime prospects of peo¬
ple on the two extremes of the distribution
of intelligence should be very different. And
they are. In this section we take a look at
the careers of some people who are at the
upper and lower ends of the intelligence
distribution.
It is important to be clear that we will
not be looking at people who are conven¬
tionally labeled “geniuses” or at people who
are mentally handicapped to the point that
they cannot function in our society without
special help. There are reasons for avoiding
these extremes.
The term “genius” is usually applied to
people who have accomplished great things.
The consensus of people who have stud¬
ied genius is that extraordinary accomplish¬
ment in any field requires some talent, but
56 Murray, 2008.
WHAT USE IS INTELLIGENCE?
343
also a great deal of motivation, and very-
hard work. The social support network must
be right. Howard Gardner makes this point
in his excellent study of extreme creators,
in such varied fields as physics (Einstein),
writing (T. S. Eliot), politics (Gandhi),
and art (Picasso).57 Gardner's subjects were
geniuses in any sense of the word. They
were all very bright. They also had a single-
minded sense of purpose and a social net¬
work of people who were willing to support
their single-minded efforts. In addition, the
times must be right. An unknown Sume¬
rian genius invented the wheel around 3500
BCE. His or her invention spread rapidly
through the ancient world, for it greatly
improved the utility of oxen and horses.
Perhaps 2,500 years later, and completely
independently, an equally unknown Aztec
invented wheels for children’s toys. The
idea never went further, for the Aztecs
had no beasts of burden. Cognitive traits
undoubtedly are important in the creation
of genius, so are noncognitive traits and fea¬
tures of the situation. Studying acknowl¬
edged geniuses is important in itself,58 but
it is not a good way to determine how high
intelligence affects one’s progress through
life.
At the other end of the scale, there is little
point in studying the lives of the extremely
mentally disabled, who simply cannot cope
with our society. Determining the sorts of
social support these unfortunate individu¬
als require is an important topic, but one
that is far beyond the scope of this book.
What we can do is to examine the lives of
people who fall in the "low normal” range,
roughly IQ scores of 70-85. Most of these
individuals are productive members of the
society, but seldom maximally productive
members.
Once again, we have to balance the
relative costs and benefits of retrospective
and prospective studies. There have been
numerous retrospective studies of the char¬
acteristics of high achievers, ranging from
57 Gardner, 1993b.
58 See, especially, Simonton, 1984.
artists to politicians.59 There have been
even more studies of various low-achieving
groups, such as welfare recipients and crim¬
inals. In both cases it is possible to find some
traits that seem to characterize the target
group. However, as is almost always the
case with retrospective studies, it is hard to
interpret these findings. For instance, eigh¬
teen of the first forty-four presidents of the
United States received earned degrees from
one or more of just five colleges.60 Does this
mean that if you attend one of these colleges
you have a good chance of becoming pres¬
ident? Hardly. Only a miniscule fraction of
the graduates of these colleges attained the
presidency.
At the other end of the social scale, it
has been estimated that about one in every
six homeless persons has either schizophre¬
nia or manic-depressive psychosis. Does
this mean that a person who suffers from
either of these diseases has roughly one
chance in six of becoming homeless? Hardly.
Approximately 3% of the US population,
roughly nine million people, suffers from
one of these two diseases. Some 200,000 are
both homeless and are either schizophrenic
or suffer from bipolar disorder. The chances
are roughly one in forty-five, not one in
six, of a mentally ill person becoming
homeless.
The best way to determine what happens
to intellectually gifted or below-normal indi¬
viduals is to start with a group of (gifted)
(below-normal) persons and follow that
group through some portion of their lives, in
a prospective manner, rather than attempt¬
ing a retrospective study of people who have
had a particular life outcome.
We will look first at a study of the gifted,
and then examine the low-normal group. In
each case my discussion will use as illustra¬
tion the results of one or two large studies.
59 See, Simonton, 1984, and Gardner, 1993b
60 The colleges are Harvard (six), the College of
William and Mary (four), Yale (four), Princeton
and the US Military Academy (two each). The list
includes George Washington, who received a sur¬
veyor’s certificate from William and Mary. This was
his only post-secondary education. I have assigned
George W. Bush to Yale, where he received a B.A.
He also received an MBA from Harvard.
344
HUMAN INTELLIGENCE
Table 10.6. Percentages of men and women in the Terman study who attained various
levels of education, compared to educational achievements of their cohort
Men in Terman Women in General General Public,
Group Study Terman Study Public, Men Women
College graduates 70 67 10 6
Graduate studies given 56 33 19 Unknown
graduation from college
Doctorates 14 4 22
Source: Data from Terman & Oden, 1959.
10.4.1. The Gifted I: The Quiz Kids
and the Termites
During the 1940s and 1950s there was a pop¬
ular radio program called The Quiz Kids. A
panel of very knowledgeable six- to sixteen-
year-old children answered questions that
required anything from an ability to do
rapid mental calculations to knowing rather
obscure scientific and historical facts. Their
performance was impressive. Some were
said to have IQs of 200, but that was
apparently a score based on the old mental
age/chronological age calculation. A more
realistic estimate is that the IQ scores ranged
in the 14^1604- range, roughly the top one
in one thousand.
About thirty years later one of the former
Quiz Kids, who had become a professor,
located a number of them to see how they
were doing.61 The commonest answer was
generally quite well, thank you. An inordi¬
nate number of them had followed academic
professions. The others were mostly in pro¬
fessional fields. One had received the Nobel
Prize in Medicine and Physiology.62 To be
sure, not every one of the Quiz Kids had
done well, and some had rather unhappy
lives. But, on the whole, they were suc¬
cesses.
The Quiz Kids hardly represents a sci¬
entific study. Candidates were recruited
rather informally from the general Chicago
area. I am sure the selection of candidates
61 Feldman, 1982.
62 James Watson, for the discovery of the structure of
DNA.
depended on both the child’s apparent intel¬
ligence and the radio show producers’ judg¬
ment about how appealing the child would
be on the radio stage. Fortunately, more for¬
mal studies have been done.
The Quiz Kid idea was a popularized ver¬
sion of a well-known study that had been
(and was being) carried out by Louis Ter¬
man, the Stanford University professor who
introduced the Binet tests into the United
States. In the early 1920s Terman asked
teachers in California schools to nominate
exceptionally bright children for a long-term
study. The children were then given IQ
tests, and those who scored 140 or above (one
in a thousand) were invited to participate.
Eventually 1,528 students were enrolled. The
study continued after Terman’ s death in
1956. Eventually the “termites,” as they were
sometimes called, were followed into their
seventies.63
The results were clear in one way, and
difficult to interpret in another. Terman ’s
participants were born around 1910-20, so
the cultural aspects of their time have to be
kept in mind. In spite of living through a
depression and a world war, they did excep¬
tionally well. A few statistics from the last
study in which Terman himself participated,
at which time the “termites” were in their
fifties, shows what had happened.64
By the 1950s virtually all of the people
in the study had completed their education.
Table 10.6 provides a comparison of their
63 Holahan & Sears, 1995.
64 Terman & Oden, 1959.
WHAT USE IS INTELLIGENCE?
345
Table 10.7. Family income distribution of Terman study participants in
the 1950s, compared to “urban white families" at that time. Income
figures are in 1950s dollars. The study participants earned much more
than the base rate established for similar families.
Group
" Urban White Families ”
Terman Participants
Income > $15,000
1%
30%
$15,000 > income > $5000
36%
64%
$5000 > income
63%
6%
Source: Data from Terman & Oden, 1959.
achievements compared to those born in
roughly the same cohort. Clearly the gifted
had much greater educational attainment
than was typical of the time. The college
graduation rate for the gifted, who went to
college during the Great Depression, was
higher than the general graduation rates at
the start of the twenty-first century! This
was true for both men and women.
Over 80% of the men in the study fol¬
lowed professional or business careers. As
was typical of the times, many of the women
became homemakers. Table 10.7 contrasts
the family incomes of the study group to
a group that Terman referred to as “urban
white families.” This was an appropriate
comparison group, for the participants were
themselves predominantly urban and white.
The final follow-up of this group, when
they were in their seventies, reinforced the
picture. The “termites” had achieved high
educational levels and had had successful
careers. They were healthy and satisfied.
Their marriage rates were high, compared
to their cohorts, and their divorce rates were
low. The incidence of severe mental ill¬
nesses, alcoholism, and other types of social
dysfunction was similarly low. The study
gave no support whatsoever to the stereo¬
type of the sickly, neurotic genius. Nor, I
add, has any other study of the gifted.
To what extent was the success of the
termites due to their intelligence? Here the
answer is not so clear. Terman’s work has
been criticized on three grounds. The most
serious is that reliance on teacher reports and
personal contacts biased the study toward
the selection of upper-middle-class urban
Whites. The second is that Terman actively
interfered in the lives of the participants,
through interviews and mail contacts. The
third is that the participants were not really
geniuses.
Terman’s selection methods were biased
toward selecting children from high SES,
White homes. On the average, although not
in all cases, Terman’s participants benefited
from strong social support, such as having
families who could support them during
their college years, and having the social con¬
tacts that facilitated success. In hindsight, it
would have been nice if Terman had also
followed children from a comparable SES
group who did not have unusually high IQ
scores. This would have greatly increased
the cost of the study. The disparity between
success rates in Terman’s group, compared
to base rates in similar social groups, is so
great that I do not think that the bias toward
upper SES participants could have entirely
accounted for the results.
Terman’s selection methods were biased
against identifying highly intelligent chil¬
dren in low SES families or in minority
groups. The bias against minorities was a
side effect of the bias toward selecting chil¬
dren from families with relatively high SES,
because in California in the 1920s the dif¬
ference in SES between Whites and Blacks
or Latinos was much greater than it is
today. There may have been many talented
individuals who should have been in Ter¬
man’s group but were not selected. How¬
ever, Terman never set out to study all,
or even a representative group, of gifted
schoolchildren in California. His intention
346
HUMAN INTELLIGENCE
was to find some gifted students and follow
them literally throughout life. He did this.
What the distribution of gifted is in the gen¬
eral population is a different question.
Would the conclusions have been any dif¬
ferent had there been an aggressive attempt
to recruit bright Black and Latino children?
We will never know.
10.4.2. The Gifted II: The Study of
Mathematically Precocious Youth
and Related Studies
We now turn to a more modern, much
larger study that is every bit as ambitious as
Terman’s was, but has a more clearly defined
recruitment procedure.
In 1971 Julian Stanley, a professor
at Johns Hopkins University, began the
Study of Mathematically Precocious Youth
(SMPY].65 Students in middle schools (then
called “junior high schools”] were urged to
take the SAT when they were twelve or
thirteen years old. Recall that the SAT is
designed for students in the third or fourth
year of high school, at age sixteen to eigh¬
teen. Stanley initially focused on mathemat¬
ical precocity, but in 1973 he began to study
verbal precocity as well.
A two-tiered selection procedure was
used. The SMPY researchers identified those
twelve- to thirteen-year-old students who
had scored in the top 3% on standardized
tests that had already been given, as part of
their school's normal assessment program,
and asked them to take the SAT. Students
who scored 500 or higher on either the
mathematics or verbal (SAT-M or SAT-V]
portion of the test were asked to partici¬
pate in the main study. This score would
put a twelve- or thirteen-year-old student in
the top half of seventeen-year-olds. Such a
level of accomplishment on the SAT-M is
an impressive accomplishment, because this
test evaluates proficiency in mathematical
topics that are often not covered until late
middle or high school. The middle school
students either had to have studied these
topics outside of school or had to apply
65 Stanley, 1996; Brody & Blackburn, 1996.
general reasoning to solve the problems on
the test. The researchers regarded the stu¬
dents who scored 500 or above to be in
the top one-half of one percent (1 in 200)
in their age group, while the students who
scored 700 or higher were believed to be
in the top one-hundredth of one percent
(1 in 10,000]. The design of the study thus
permits comparison between the bright and
the extremely bright.
Stanley died in 2005. The SMPY study
has been carried on by Stanley's colleagues,
Camilla Benbow and David Lubinski, at
Vanderbilt University. Their intent is to
make SMPY a fifty-year study of the careers
of people identified as talented in their
early adolescence. (Stanley's first partici¬
pants were born in the late 1950s and early
1960s.] In addition to observing the behav¬
ior of talented students, Stanley and his col¬
leagues were interested in the nurturing of
exceptional talent, especially in mathemat¬
ics. Therefore, the SMPY has included a
teaching component, in which participants
are enrolled in intensive summer programs.
The data that has been gathered at this
time (2010], which roughly carries the partic¬
ipants through their thirties, provides insight
into several aspects of the development
of the gifted. These include reactions to
academic environments, academic achieve¬
ment, and accomplishments by early middle
age. 66
The students took to intensive academic
instruction like ducks to water. The SMPY
students could assimilate a year's high school
course work in about three weeks of inten¬
sive study. This is consistent with other
reports, indicating that when special instruc¬
tion is offered it helps everyone, but it
helps the talented students the most.67 I
would add a caveat to this. The state¬
ment is true unless the special instruction
is highly structured and specifically devel¬
oped for a low- ability group. In this case the
high-ability group’s accomplishments may
66 The data cited here is largely taken from Lubinski
and Benbow' s (2006) review of the status of the
project thirty-five years after the initial enrollment
of students.
67 Ceci & Papierno, 2005.
WHAT USE IS INTELLIGENCE?
■ BA/BS
■ MA/MS
□ Doc.
Cohort-Gender - Number of Cases
Figure 10.11. Educational attainment of participants in the SMPY,
by age thirty-three. Members of cohorts 1 and 2 attained scores of
500 or better on the SAT-M by age thirteen. Members of cohort 3
attained scores of 700 or better. For reference, in the general
population in the US as of 2004, 27% of the population twenty- five
or older had received bachelor’s degrees, 9% had master’s degrees,
and 3% held a doctorate of some type, including the Ph.D., M.D.,
L.L.D., and E.D. degrees. Data from Lubinski & Benbow, 2006.
347
actually deteriorate due to loss of interest
and motivation.68
The academic achievements of the group
are staggering. Figure 10.11 shows the level of
academic achievement obtained by SMPY
participants who had achieved scores of
either 500 or greater or 700 or greater on the
SAT-M. Lubinski and Benbow have pointed
out that it is, to say the least, interest¬
ing that a two-hour test taken at age thir¬
teen or younger can predict that the "risk,”
if you will, of obtaining a doctorate has
risen from three in one hundred, the U.S.
national rate, to one in two (cohort 3 in
Figure 10.11].
What about on beyond schooling? At the
time of this writing (2010) the SMPY par¬
ticipants are less than fifty years old. Nev¬
ertheless, their accomplishments are spec¬
tacular. Because many of the participants
went into science, engineering, and related
careers, a particularly compelling compari¬
son contrasts the SMPY participants to grad¬
uate students of a similar age who had
attended highly ranked university programs
in science or engineering. Figure 10.12 shows
68 Snow, 1982, 1996.
this comparison for a distinctly academic cri¬
terion, employment at highly ranked uni¬
versities, and a distinctly nonacademic cri¬
terion, how much money the individual
was making. By age thirty-three approxi¬
mately 8% of the "over 700” men were earn¬
ing more than $250,000 as individuals. In
2005 (the time of the survey] approximately
1.5% of all US households earned more than
$250,000.
The SMPY was initially motivated by a
desire to study the life histories of ado¬
lescents who showed exceptional promise
in mathematics. As the study progressed it
extended its reach to the evaluation of peo¬
ple with high SAT-V scores. This made it
possible to study people whose talents were
tilted toward either mathematical or ver¬
bal skills. It should be remembered, though,
that people with very high verbal scores will
probably have above-average mathematics
scores, and vice versa.
The tilt definitely affected the type of
contribution that was made. SMPY partic¬
ipants whose strongest test scores were on
the SAT-V tended to major in the social sci¬
ences and humanities; those with high SAT-
M scores chose the sciences and engineer¬
ing fields. This difference held true both for
348
HUMAN INTELLIGENCE
Males Females Males Females Males Females Males Females
Graduate Talent Graduate Talent
Students Search Students Search
Figure 10.12. The SMPY participants compared to graduate
students of the same age. A comparison of the accomplishments,
by age thirty-three, of SMPY participants who had scored 700 or
better on either the SAT-V or SAT-M, at age thirteen, to the
accomplishments of graduate students from highly ranked U.S.
university programs. From Lubinski and Benbow, 2006, Figure 3.
choice of major and for later professional
work.69
The mathematically precocious youth did
not rest on their laurels. They were willing
to work more hours than the graduate stu¬
dents to whom they were compared - even
though mathematics and science graduate
students, all in highly rated programs, are
a notoriously hard-working group. Some¬
what more men than women qualified for
the program. The disparity between men
and women increased sharply as a func¬
tion of the SAT-M score. This can be seen
from an examination of Figure 10.11. The
male/female ratio in cohorts I and II, which
were in the top 1/200 group, is 1.6 to one.
The male/female ratio in cohort III, the 1 in
10,000 group, was 8.7 to one.
The SMPY participants tended to come
from small families of relatively high SES.
Their parents were much more likely to
have college degrees, including advanced
degrees, than would be the case for ran¬
domly selected students. These two relation¬
ships are not independent, for family size is
negatively correlated with SES.
69 Park, Lubinski, & Benbow, 2007.
Only about 1% of the SMPY participants
were African American or Latino. This is
comparable to the figure in Terman’s study,
although the percentage of African Amer¬
ican and Latino students in the schools
had increased markedly between the SMPY
recruitment and Terman’s recruitment, fifty
years earlier.
Thirty-three percent of the participants
in the SMPY and related programs are
Asian.70 Nationally, Asians constitute about
4% of the population. It has been claimed
that Asians have a greater genetic potential
for intelligence, and especially for mathe¬
matical reasoning. This claim is somewhat
controversial.71 However, genetic potential
is unlikely to have been a cause for the
overrepresentations of Asians. Eighty per¬
cent of the Asian participants had parents
who had been born and educated outside of
the United States. Were the contrast to be
due to genetics, one would have expected
a much higher percentage of Asians whose
70 Brody & Blackburn, 1996. Unfortunately these
authors did not provide figures for the various Asian
groups.
71 See Flynn, 1991. This topic is discussed in more detail
in Chapter 11.
WHAT USE IS INTELLIGENCE?
349
parents had been born in the US. Nor is
the contrast likely to be due to some great
deficiency in US schooling, for most of the
non-Asian participants had been educated
in the US school system. This strongly sug¬
gests that environmental factors influenc¬
ing the home life of recent Asian immi¬
grants were a major factor in the children’s
developing interests and skills in academic
pursuits.
The Terman study and the SMPY are not
the only studies of the gifted, but they are
perhaps the largest, and their results are rep¬
resentative. People who do well on cognitive
tests in their late childhood or early teens
have quite bright prospects. High scorers
tend to come from fairly high SES fami¬
lies, so some of their success may be due
to the advantages of privilege. However, the
accomplishments of the gifted are too sub¬
stantial to support a claim that advantage is
all they have. Their own ability counts for a
great deal.
10.4.3. Developing the Gifted
A variety of acceleration programs have
been developed to assist the gifted in reach¬
ing their potential. These vary from spe¬
cial summer courses, as in the case of the
SMPY, to provision of accelerated tracks in
public high schools and offering children
as young as fourteen early admissions to
universities. Gifted children perform very
well in such programs. They also report
enjoying them. There is little evidence for
widespread social maladjustment, although
participants in the early college entrance
programs report preferring the company of
their equally gifted age-mates to that of
the considerably older college students. One
particularly telling comment was made by
Nancy Robinson, a professor at the Uni¬
versity of Washington who, together with
her husband, Halbert, developed an early
entrance program at that university. She
noted that gifted students who came from
the public schools benefitted from a pre¬
training program that prepared them for the
pace of instruction at a major university.
Why? Because they needed to develop study
habits] They had been able to get by in their
regular schools without exercising the disci¬
pline needed when they were thrown into
the far less supportive atmosphere of uni¬
versity instruction.72
10.4.4. ^ Comment on Criticisms
of the Concept of the Gifted
On occasion the results of the studies of the
gifted have been criticized in ways that I
think are unfair or irrelevant. For instance,
I have heard the Terman study criticized
for not having identified a Nobel laure¬
ate! There are also widespread, although as
far as I know undocumented, stories that
two men who did subsequently win Nobel
Prizes in science were overlooked during
the selection of participants. Such criticisms
set an unrealistic standard, by asking the
researchers to predict a one in a million
event, while ignoring major trends in the
data.
Another criticism is a claim that very high
scores on cognitive tests do not predict any¬
thing. This belief is extremely widespread,
even among psychologists.73 It is false. In
the SMPY there is a substantial difference
between the accomplishments of the top 1
in 200 among test scorers and the accom¬
plishments of the top 1 in 10,000. Similar
findings were noticed by Terman for people
with IQs over 170 as children.
10.4.5. The Prospects for Individuals
ivith Low Test Scores
What about the other side of the coin, peo¬
ple in the low normal intelligence range,
which I shall define as those whose IQs lie
between 70 and 90?
People in the low intelligence range are
not automatically candidates for assisted liv¬
ing or other institutional programs. How¬
ever, more people in the low intelligence
range are found in welfare and prison/jail
72 See Cronbach, 1996, and Robinson, 1996, for elabo¬
ration on these points.
73 See Muller et al., 2005, and Vasquez & Jones, 2006,
for examples of such assertions.
HUMAN INTELLIGENCE
35°
populations than would be found if the
welfare and prison/jail populations were
selected randomly from the population. This
does not mean that the majority of low intel¬
ligence individuals are headed for some form
of institutional control. It simply means that
they are at a greater risk of having these bad
things happen than is the case for a randomly
chosen member of the population.
As in the case of the gifted, the best way
to understand the issues facing people in the
low intelligence range is to examine prospec¬
tive studies, in which individuals in the low
intelligence range are first identified, and
then, hopefully along with a control group of
average and above-average individuals, are
followed for some time. Here are three such
studies.
In Herrnstein and Murray’s “Bell Curve”
study74 of the NLSY79 panel, teenage men
and women took the AFQT in their mid
to late teens.75 The Department of Defense
uses these scores to classify people into cat¬
egories I-V, with categories IV and V cov¬
ering the range of scores below 90. About
45% of the NLSY79 men and women in
categories IV and V failed to obtain either
a high school diploma or a general educa¬
tion degree (GED). The base rate for the
entire NLSY79 panel was 9%. Measures of
workplace performance were similarly low.
Herrnstein and Murray concluded from this
that intelligence, as measured as a teenager,
is a major predictor of income as a young
adult. Other analyses of the same data76
have agreed that intelligence is indeed a pre¬
dictor of income, but that ethnic status, gen¬
der, and location in the country also have to
be considered.
If we look at analyses of the intellectual
demands of the workplace, this is not sur¬
prising. Look back to Figure 10.10, which
plots wages against imputed intelligence
requirements for over 800 jobs. Income is
fairly flat over the wide range of occupations
that have low imputed intelligence require¬
ments. The figure also shows that there is
74 Herrnstein & Murray, 1994.
75 See Chapter 2 for a discussion of the ASVAB and
the AFQT.
76 See, e.g. the analyses in Devlin et al., 1998.
a considerable spread between the mean
and ninetieth percentile of earnings within
occupations. To the extent that intelligence
determines within-occupation earnings, as
we have seen that it does (section 10.3), one
would expect people with low scores to earn
less. And they do.
The picture is somewhat bleaker with
respect to welfare. In this case we have
to make a distinction by ethnicity, for wel¬
fare rates vary markedly with ethnic status.
Herrnstein and Murray found that the rate at
which White women in the low intelligence
range had received Aid for Dependent Chil¬
dren support was better than four times the
rate in the NLSY79 sample as a whole.
A more detailed picture of what hap¬
pens to the low intelligence group in the
workplace can be obtained by examining a
Department of Defense study of how low
intelligence males fared in military service,
in a setting where working conditions are
more precisely defined and recorded than
they are in the civilian workforce. The study,
Project 100,000, conducted in the late 1960s
at the behest of Secretary of Defense Robert
McNamara, was motivated by an important
policy issue.
The US military normally does not recruit
Category V individuals, and is limited, by
law, to recruiting a fixed percentage of Cat¬
egory IV soldiers. The argument for the pol¬
icy is that it does not make sense to go to
the added expense of training and supervis¬
ing personnel who perform at a low level.
On the other hand, the military forces have
to have a certain number of recruits each
year. If the category IV designation is a false
indicator of military performance, exclud¬
ing these men and women would amount to
ignoring a potential recruiting population.
In Project 100,000 approximately 100,000
Category IV men (IQ range roughly 80-90)
were enlisted outside of the normal chan¬
nels. Their military careers were compared
to those of a control group of enlisted ser¬
vicemen who matched them in age and edu¬
cational status prior to entry, and who had
met normal recruitment standards. The con¬
trol group underrepresented the higher lev¬
els of AFQT scores (I and II) compared to
WHAT USE IS INTELLIGENCE?
351
Table 10.8. Attrition rates during basic training (percentages) for
Project 100,000 participants and for the service as a whole, broken
down by military branch.
Branch of Sendee
Project 100,000 Enlistees
Overall Service Rate
Army
37
2.0
Marines
11.1
4-4
Navy
8.6
2.8
Air Force
9.2
3-o
Note: Data from Sticht et al., 1987. Figures are for the 1969-72 period, during
the Vietnam War.
the percentages found in the general popula¬
tion, for there were no officers in the study.77
In civilian terms, Project 100,000 exam¬
ined the workplace performance of peo¬
ple involved in blue-collar and lower-level
white-collar occupations, excluding man¬
agerial positions above the foreman level,
and excluding professional occupations.
In all military services the first thing that
enlistees do is to go through recruit training
or, as it is known in the Navy and Marines,
boot camp. The ostensible goal of recruit
training is to inform the enlistees about ser¬
vice customs and to provide them with a
taste of the life they can expect in the future.
This taste (and the service life that follows)
varies greatly depending upon the service;
the Army and Marines envisage a different
life for recruits than do the Navy and the
Air Force, and the Navy and Air Force dif¬
fer from each other. Boot camp has a sec¬
ond, less announced purpose. It serves as a
screening device to see which enlistees have
the adaptability to change from civilian to a
more disciplined military life.
Table 10.8 shows the attrition rate from
boot camp during the period of the Project
100,000 study, roughly 1969-71. Attrition
rates were higher for Project 100,000 than
for the control group in every service. Attri¬
tion rates also varied considerably across
services. This may be because of both dif¬
ferential recruitment by the services and dif¬
ferences in basic training itself. Basic military
training, strictly construed, requires roughly
a fourth-grade reading ability. Because the
77 Sticht et al., 1987.
Navy and the Air Force have more tech¬
nical billets than the other services, higher
reading and mathematics requirements may
have been in force in those services. These
requirements would be especially hard on
the Project 100,000 servicemen, compared
to members of the control group. By con¬
trast, the Army and the Marines place more
stress on determining a recruit’s ability to
follow instructions in physically demanding
situations.
Looking at the other side of the coin,
the vast majority of the Project 100,000 ser¬
vicemen did complete basic training. From
the viewpoint of the services, a rigid insis¬
tence on conventional standards would have
excluded over 90,000 trainable enlistees.
That would have been a serious loss.
Modern military services are a microcosm
of society. They contain some positions,
such as infantryman, that are clearly unique
to warfare. They also contain many posi¬
tions that either have exact civilian counter¬
parts, such as clerk or electronic technician,
or have close analogs in civilian life, espe¬
cially outside of combat. We have already
seen, from the analysis of civilian occupa¬
tions in section 10.3, that occupations vary
a good deal in their demands for intelli¬
gence. Where did the Project 100,000 ser¬
vicemen wind up during their first term of
service?
Figure 10.13 shows that Project 100,000
servicemen were much more likely to be
assigned to nontechnical, nonspecialized
jobs than were normally enlisted servicemen
matched for pre-service education and eth¬
nic status. However, the evidence that this is
352
HUMAN INTELLIGENCE
50
■o 45
0) 40
£ 35
8 30
< 25
c 20
■ Project 100,000
■ Control Group
Figure 10.13. Assignment of Project 100,000 and a matched control
group to different classes of military occupations during their first
term of enlistment. Based on data reported by Sticht et al.; 1987.
due to their demonstrated abilities (or lack
of them) after testing is not as strong as it
could be. Assignments to technical billets
are made based partly upon observations
and interviews during recruit training, and
partly upon a serviceman's ASVAB scores,
obtained prior to entry. This means that the
low scores of the Project 100,000 servicemen
may have influenced their assignment.
From the viewpoint of a scientific study,
this was a flaw in the design. However, one
can hardly fault the services for using the
scores to make assignments in ways believed
to minimize training costs.
As in the case of the data on attrition,
one can see two different things in the data
on occupations. On the one hand, it shows
(not surprisingly) that low intelligence men
tended not to be assigned to technical and
administrative occupations. On the other,
they filled roles that are vital to the military
mission.
What happened after the servicemen
completed recruit training? Figure 10.14
compares the career progress of Project
100,000 and control servicemen, in terms of
rank attained through their first enlistment
and, for those that stayed in the service, their
status in 1983, which would have been from
twelve to fourteen years after initial enlist¬
ment. The picture is similar to the earlier
comparisons. The Project 100,000 service¬
men were generally not “top of the line,"
rapidly promoted military personnel. On the
other hand, they did show progress in their
careers.
Project 100,000 was replicated . . . acci¬
dentally. During the 1980s a technical mis¬
take was made in determining the norms
for a revised version of the ASVAB. Before
the mistake was discovered several thousand
normally unqualified Category IV soldiers
were enlisted into the Army. The careers of
these soldiers have been followed, and by
and large the results are the same as those
obtained in the better-controlled Project
100,000 study.78 There would be little point
in repeating the statistics. However, I will
recount a hopefully informative anecdote to
enliven the statistics.
While serving as a consultant to a research
project studying the accidentally recruited
Category IV soldiers, I interviewed a lieu¬
tenant colonel who had commanded an
armored battalion containing a high per¬
centage of Category IV soldiers. The colonel
believed that category IV soldiers performed
well when doing clearly defined tasks, even
if they were quite detailed. He offered as an
example the task of replacing the power unit
on a tank. This task is done in over a dozen
well-defined steps, always done in the same
sequence. After training, Category IV sol¬
diers could replace the power unit.
78 Sticht et al., 1987, discusses the renorming prob¬
lem and the resulting performance of the soldiers
involved in some detail.
WHAT USE IS INTELLIGENCE?
353
E1-E2 E3 E4 E5 E6 E7 E8
■ Project 100,000 1st
Enlistment
1 Control 1st Enlistment
* Project 100,000, 1983
□ Control 1983
Figure 10.14. The advancement in rank of Project 100,000
personnel. The percentage of servicemen in the Project 100,000 and
control groups who had reached different enlisted pay grades either
during their first enlistment or after approximately thirteen years of
service. Pay grades Ei and E2 were the grades typically assigned
immediately upon completion of recruit service (e.g., private or
private first class in the Army and Marines). Pay grade E4 (sergeant
in the Army or Marines) was considered satisfactory service during
the first enlistment. The higher pay grades represent staff
noncommissioned officers (NCOs) or, in the Navy, Petty Officers.
Grade E8, Master Sergeant or Chief Petty Officer, was a position
with considerable prestige and responsibility.
More generally, the colonel believed that
Category IV soldiers could be trained to do
tasks where the instructions were “do this,
then do that, then the other thing
The colonel thought that Category IV sol¬
diers had trouble with tasks that are defined
by the end to be accomplished, rather than
by the steps to be taken. He offered as an
example the task of recalibrating a gun sight
that has been knocked out of alignment with
the barrel of a tank’s cannon. The instruc¬
tions for this task began, “Find a way to
fix the sight rigidly on the tank’s hull.” This
instruction defines the goal to be accom¬
plished and leaves it up to the soldier to find
a way of accomplishing the goal.
If we place this in a more psychologi¬
cal framework - for the colonel certainly
did not use such words - the Category IV
soldier could learn what to do in a well-
defined situation but had trouble anticipat¬
ing what would happen if he took a particu¬
lar sort of action in a more poorly defined
situation. This sort of deficit appears to
be characteristic of low intelligence people
in schools, the military, and the civilian
workplace. Depending upon whether you
approach intelligence from the viewpoint
of the psychometrician, the information¬
processing psychologist, or the cognitive
neuroscientist, you can say that people
with low intelligence have trouble with
tasks that have a high demand for g, tasks
that involve the working memory/executive
function class of behaviors, or tasks that
place demands on the forebrain-cingulate
cortex circuit. All three statements amount
to the same thing.
10.5. A Concluding Comment on
Intelligence and the Workplace
Someone who says that intelligence, as
assessed by standard cognitive tests, is irrel¬
evant to performance in either academia
or the workplace is simply wrong. So is
someone who says that intelligence does not
amount to very much, compared to a vari¬
ety of personality characteristics. The data
354
HUMAN INTELLIGENCE
both within and across occupations show
that measures of intelligence are among the
best, and most often the best, predictors of
academic and occupational success. On the
other hand, no one has claimed that cogni¬
tive tests are perfect predictors. Predictive
validity correlations are in the .4-6 range,
which is much better prediction than can be
achieved with any personality measure, but
is still far from perfect.
The same message can be extracted from
studies of extremes. On the whole, people
who have high intelligence test scores do
quite well. On the average the gifted get bet¬
ter jobs and make more money. The Terman
study, and all studies of the gifted afterward,
gave the lie to the stereotype of a gifted per¬
son as a neurotic introvert with health prob¬
lems. The contrast between the SMPY 1 in
200 and 1 in 10,000 groups shows that the
tests have predictive power at very high lev¬
els, directly contradicting statements to the
contrary by psychologists who do not, them¬
selves, study individual differences.
Here are some documented statistics,
but on a very small sample, so the report
falls somewhere between a scientific fact
and an anecdote. In the 1960s the National
Aeronautic and Space Agency conducted
an intensive screening of volunteers to
become the first American astronauts, the
MERCURY and GEMENI programs. The
selected candidates had WAIS IQs averaging
135. Unselected candidates had IQs averag¬
ing 131. A control group of comparably aged
aviators averaged 118. Being intelligent is part
of having the right stuff.79
At the opposite end of the distribution,
people in the low intelligence range gener¬
ally have difficulty with school, especially
if they are placed on an academic track,
and usually take occupations that do not
make high cognitive demands. They earn
less than people in the normal-high intel¬
ligence ranges and are more likely to require
some form of welfare assistance.
It cannot be stressed too strongly, though,
that these are trends. Every study of the
extremes comes up with exceptions. There
79 Santy, 1994, p. 276.
are people who do quite well although they
had modest test scores, and there are stun¬
ning examples of people with high scores
who never live up to their promise.
Given these facts, why is there a
widespread belief that intelligence does not
count for very much in life? I think, but can¬
not prove, that several factors are involved.
One factor is a failure of unrealistic
expectation. We may expect the gifted per¬
son to be a casual genius who can solve diffi¬
cult problems without much care or effort.
That is not the case. Two of the personal¬
ity characteristics of the gifted are that they
generally enjoy their work, and that they
work very hard at it. The brilliant genius
who, without training, knows at a glance
the answer to difficult problems in math¬
ematics, physics, or what have you is a very
rare bird. 80
People without statistical training have a
hard time grasping the concept of something
that increases the probability of an event
but does not establish its certainty. Thus if
we can think of examples of people with
high test scores doing stupid things, or peo¬
ple with low test scores doing good things,
that is taken as proof that the predictors do
not work. In the newspaper business “Man
bites dog” is news, “Dog bites man” is not.
The unusual is publicized and sticks in our
minds. The prosaic does not.
We may have an unrealistic idea about
the extent to which personal characteristics
determine success. Large-scale social and
economic forces, and idiosyncratic imper¬
sonal events, can play a great part. The
various reports of the Terman group stress
how much these highly intelligent people
were influenced by having come of age dur¬
ing the Great Depression and then, espe¬
cially for the men, having had to deal with
World War II. The SMPY group has grown
up in times of relative peace and economic
expansion, at least until 2010. Such things
80 It is possible that a very few such individuals exist.
The best-documented case study is of the Indian
mathematician Ramanujam, who made major con¬
tributions to mathematics even though he was self-
taught. He also spent a great deal of time working
on mathematical problems.
WHAT USE IS INTELLIGENCE?
355
influence one’s success, quite outside of per¬
sonal traits. While it is true that we partly
make our own environments, we are partly
stuck with them.
People are heavily influenced by their
personal experiences. Charles Murray has
pointed out that we live in a society that is
sharply stratified by intelligence.81 College-
educated people, by and large, deal with
other college-educated people, and people
with high school educations deal with each
other. Within the restricted range of intelli¬
gence that people can observe directly, other
variables may account for more variation
in performance than intelligence does. It is
only when we step back and look at the big
picture that the importance of intelligence
becomes clear.
My final speculation is that some peo¬
ple, for understandable reasons, do not want
to find that intelligence has an effect upon
success.
Many people in post-industrial society,
and especially in post-industrial American
society, hold a belief and have two attitudes
that, combined, provide a motivation for
rejecting intelligence as a (partial) explana¬
tion of workplace success. The belief is that
intelligence is something that is more or less
fixed for life. As was discussed in Chapter 9,
and will be discussed further in the section
on aging in Chapter 11, this is a false belief -
especially about intelligence in the concep¬
tual sense of the problems that people can
solve, rather than in the narrower sense of
a test score. But that is not what people
think. When one combines a belief in the
permanence of intelligence with an attitude
of distrust of elites, it becomes almost nec¬
essary to argue that intelligence is of little
relevance in life, for to do otherwise can be
81 Murray, 2008.
seen as an affirmation of the appropriateness
of rewarding an elite class of thinkers.
The second attitude is quite different. It
has to do with a sincere desire for equal
opportunity for all.
There are marked differences in the typ¬
ical cognitive test scores obtained by mem¬
bers of different racial and ethnic groups.
There are much smaller, rather complex dif¬
ferences in test scores between men and
women. Up to the middle of the twentieth
century these differences were used to jus¬
tify varying degrees of segregation of minor¬
ity groups, in areas including admission to
universities and the granting of specialized
degrees, and to place less strict, but still
important, restrictions on women’s oppor¬
tunities. Since that time overt discrimina¬
tion has virtually stopped, but there are peo¬
ple who have used group differences in test
scores as evidence for the proposition that
group differences in educational and profes¬
sional achievement are largely due to group
differences in intelligence.82 To the extent
that this conclusion is correct, provision of
equal opportunity for all groups in society
will not produce an equal distribution of
social and economic rewards across groups.
Many people who are deeply commit¬
ted to social equality find such a conclusion
offensive. It is difficult for them to argue
about the fact of differential distribution of
test scores. Therefore, they deny the rel¬
evance of scores to social outcomes. This
denial cannot be maintained.
The discussion of racial and ethnic differ¬
ences in intelligence, and male: female differ¬
ences in intelligence, raises extremely com¬
plex issues. We take them up in the next
chapter.
82 See, for instance, Murray, 2005; Rushton & Jensen,
2005; and Lynn and Vanhanen, 2002, 2006.
CHAPTER 1 1
The Demography of Intelligence
By nature men are nearly alike. By practice
they get to be far apart.
- Confucius (attribution in
Bartlett's Familiar Quotations,
15th ed., 1980)
People are not all the same. There are young
people, old people, and people in between.
Nevertheless, we do make distinctions, rang¬
ing from binary decisions about the right
to vote to the minimum age at which one
qualifies for a pension. Many of these dis¬
tinctions are based on the underlying pre¬
sumption that intelligence grows, and then
declines. There are mandatory retirement
age requirements for air traffic controllers
and commercial aviators, largely because of
our beliefs about changes in their cogni¬
tive capabilities. In the United States federal
judges are appointed for life. Many of the
individual states have mandatory age limits
for judges.
There are men and women, a sharper bio¬
logical distinction than young and old. His¬
torically, many societies have assumed that
men and women have different cognitive
capacities. In contemporary post-industrial
societies there is a presumption of equal¬
ity. But does “equality” mean “identity?”
Should we regard a low incidence of women
in corporate law practice with suspicion?
Should the acceptable percentage of women
in corporate law practice be the same as the
acceptable percentage of women working as
helicopter pilots?
People are identified with, and self-
identify with, various racial and ethnic
groups. Do these groups differ in intelli¬
gence? To some people the answer is obvi¬
ous - they do. Other people believe that
even the suggestion of a difference is evi¬
dence that the speaker is prejudiced. World¬
wide, we are organized into nations. Do the
people of the world all have the same intel¬
ligence? Herodotus, the father of history,
thought that easy climates produce people
with soft minds and little resolve. Richard
Lynn, a modern professor at the Univer¬
sity of Ulster, has given new life to this
argument.1
1 Lynn, 2006.
THE DEMOGRAPHY OF INTELLIGENCE
357
And identifying group differences is sel¬
dom enough. We want to know why differ¬
ences occur.
This chapter describes the scientific evi¬
dence regarding the relation between intel¬
ligence and demographic variables: age, sex,
race, and ethnicity. I shall try to present
an objective discussion of these contentious
issues.
11.1. The Issues Involved
I will not present any simple one-liner
answers to any issue involving group differ¬
ences. In my opinion none exists. A person
who makes a sweeping assertion about the
existence or cause of age, sex, racial/ethnic,
or national differences should be regarded
with extreme suspicion. Throughout I have
tried to separate discussions of scientific fact
from hypotheses and from discussions of
policy. Maintaining the distinction between
scientific findings and policy recommenda¬
tions is very important. Psychologists can,
for instance, inform lawmakers about the
facts that reaction times increase and the
ability to control attention drops as people
age. Whether or not there should be upper
age limits on automobile driving is a policy
decision.
Because the study of group differences
can be so contentious, it is a good idea to set
the ground rules by discussing some prob¬
lems of analysis and interpretation that keep
coming up in this area of investigation. The
rules fall into two categories: general prin¬
ciples and issues of interpretation. Panel 11.1
sets forth some general principles. The rest
of this section focuses on issues of interpre¬
tation.
11.1.1. Distinguishing between Cognitive
and Noncognitive Effects
Intelligence is a “can do” concept. The dis¬
tinction between “can do” and “will do” is
relevant to the study of group differences,
because when group differences in perfor¬
mance are observed it may not be clear
whether they are due to differences in intel¬
ligence or motivation.
The case of Asian-Americans in the
United States provides a good example. As
a group, Asian-Americans are represented
in higher education far out of proportion
to their frequency in the population. In
autumn of 2008 Asians made up 35% of the
student body at the University of Califor¬
nia, Berkeley. Asians constitute just over 4%
of the US population, and 12% of the pop¬
ulation of California.2 Some have attributed
the overrepresentation of Asians in selective
universities to Asian intelligence, for Asians
as a group score slightly above the popu¬
lation mean on some g-loaded intelligence
tests. This point is discussed in more detail in
section 11.3. But is this because of a biological
capacity? One also has to consider cultural
traditions and expectations about the advan¬
tages of education that may provide Asian-
Americans with more-than-average motiva¬
tion to succeed in cognitive endeavors.5
A good deal has been made of the pos¬
sibility that some people consistently score
below their capabilities on cognitive tests
and other measurements of achievement
because they just do not try. The phe¬
nomenon is by no means limited to cog¬
nition; if people have a mindset to expect
low achievement, they will not work hard to
become high achievers in almost any field.4
In the case of group differences, it has been
claimed that a particular type of negative
mindset, stereotype threat, may be one of
the reasons that certain groups, including
African Americans and women, tend not to
do well on tests of mathematical reasoning.
They believe that they are not the sort of
people who do mathematics well, so they
are willing to disengage from the hard work
of problem solving required to obtain high
scores on mathematics tests.5
2 Information provided by the University of Cali¬
fornia, Berkeley, at the website 0sr2.berkeley.edu/
twiki/bin/view/Main/Fall20o8EthnicDistribution,
and by the 2000 US Census.
3 Flynn, 1991; Sue & Okazaki, 1990.
4 Dweck, 2006.
5 Steele & Aronson, 1995, 1998.
358
HUMAN INTELLIGENCE
Panel 11.1. Some Principles for
Guiding Research on Group
Differences in Intelligence
The following rules are abridgments of
suggestions about conducting research on
group differences that my colleague Prof.
Jerry Carlson, of the University of Cali¬
fornia, Riverside, and 1 made in an article
in the journal Perspectives on Psychological
Science .* Our ideas were attacked on the
somewhat contradictory grounds that our
guidelines would unduly restrict inquiry^
and that they would encourage delving
into questions that are not scientific and
that we did not understand/ We made
a rejoinder.** 1 encourage readers inter¬
ested in this issue to look at the origi¬
nal articles. Here I simply paraphrase the
guidelines.
1. The measures must have construct
validity, that is, it should be clear why
they measure what they purport to
measure.
2. Measurements must be valid in all
groups involved. See the main text
for a discussion of what this means.
3. The fact that a score can be changed
by training is not evidence against an
inherent group difference unless the
altered score is as valid a measure as
the original score.
4. Generalization to populations
depends crucially upon the relation
between the sample and the popu¬
lation. "Any old convenient sample”
will not do. A generalization from
observations of a group difference in
college students, for instance, to a
conclusion about group differences
for people of all ages is not automat¬
ically valid.
3. Literature summations must be done
carefully, with special attention to
research results that do not con¬
form to the reviewer’s conclusions.
Complete objectivity is impossible,
but discussants in a scientific debate
should strive for this ideal.
6. Alternative hypotheses and models
should be considered.
7. The alternatives should duly repre¬
sent the original authors’ ideas. Straw
models should not be set up in order
to be knocked down. This has been
a particular problem in the study of
group differences. For instance, peo¬
ple sometimes attack the position
that group differences are either enti¬
rely genetic or entirely environmen¬
tal, whereas the real question is how
much the environment or genetic
inheritance influences the difference.
8. Heritability coefficients are measures
of the relative size of genetic and
environmental influence, and can
vary across populations.
9. When a policy recommendation is
made, one’s policy model, includ¬
ing attitudes about desirable conse¬
quences, should be stated.
10. Be willing to say "We don’t know.”
In many cases we do not know what
causes a group difference. There are
some cases, especially involving the
evolution of cognition, where it is
unlikely that we shall ever know.
Be willing to acknowledge such sit¬
uations. Acknowledging ambiguity is
not a sign of weakness, it is a sign of
intelligence!
* Hunt & Carlson, 20073,0.
* Gottfredson, 2007a.
Sternberg Sc. Grigorenko, 2007.
3 Hunt Sc Carlson, 2007b.
There is no doubt that stereotype threat
exists. The effect has been demonstrated
in laboratory studies involving racial/ethnic
groups, men and women, and aging par¬
ticipants. There is considerable controversy
over whether the phenomenon occurs when
students take a high stakes test, such as
a college entrance examination, because
THE DEMOGRAPHY OF INTELLIGENCE
359
when the stakes are high external motivating
factors may override any internal willingness
to give up. The evidence for and against
stereotype threat effects will be discussed
later. Here I bring up the problem as an
example of how behavior will depend upon
both "can do” and “will do.” Before we can
use a difference between two groups in test
scores, or any other cognitive achievement,
as evidence of a difference in underlying
intelligence, we must be sure that motiva¬
tion was equivalent in each group.
11.1.2. Recruitment Effects
It is probably true that more studies have
been carried out on college students than
on any other population. For instance, a
great deal of our information about the dif¬
ferences in cognitive abilities of men and
women is based on observed differences
between male and female college students.
Such studies are subject to recruitment effects
that limit their generality. Let us look at such
situations abstractly.
In the typical study of group differ¬
ences an investigator identifies some acces¬
sible population (e.g., college students) that
contains members of the groups of inter¬
est. Obviously, there are male and female
college students. Participants are selected
from the groups of interest within the
accessible population. Group differences are
then observed. Assuming a random selec¬
tion from the accessible population, con¬
ventional statistical methods can be used to
justify making a general statement about the
accessible population. But what about con¬
clusions about group differences in the pop¬
ulation at large? The validity of generaliza¬
tion depends upon how group members are
recruited into the accessible population.
Here is a concrete example. Two Cana¬
dian psychologists, Douglas Jackson and J.
Phillipe Rushton, found that in the popu¬
lation of people who had taken the SAT,
men had higher scores than women.6 Does
this mean that men in their late teens (the
age at which the SAT is usually taken) are
6 Jackson & Rushton, 2006.
generally more intelligent, as assessed by the
SAT, than women of the same age? The con¬
clusion would be valid if young men and
women, in the population at large, were
equally likely to take the SAT. But they
are not. Since 1980 the majority of college
students have been women. It may be that
the most intelligent 60% of women attempt
to go to college, compared to the most
intelligent 40% of men. If this is so, the
recruitment process is different for men and
women.
In such a case comparison of the SAT
scores would give a false picture of the popu¬
lation difference. If there are no male-female
differences in intelligence in the general
population, the top 60% of the female popu¬
lation would be expected to score lower, on
the average, than the top 40% of the male
population. Of course, this is an oversim¬
plification. A more sophisticated model of
male-female differences in the recruitment
process for taking the SAT has been verified.
Applying it, my colleague Tara Madhyastha
and I found that Jackson and Rushton's
results could not be used to argue that there
are differences in intelligence between men
and women in general, although they could
be used to argue that there are differences in
intelligence within the college population.7
Recruitment effects can be very large.
They affect all types of group differences
that we will consider: age differences, male-
female differences, racial/ethnic differences,
and international comparisons.
11.1.3. Establishing Causation
Students in the social sciences learn to recite,
almost as a mantra, “Correlation does not
mean causation.” The rooster may crow
before dawn, but we do not conclude that
the rooster’s crowing makes the sun rise.
In laboratory situations we avoid confus¬
ing correlation with causation by conducting
controlled experiments, where one possi¬
ble causal factor is systematically changed as
other factors are held constant, or by assign¬
ing participants randomly to experimental
7 Hunt and Madhyastha, 2008.
360
HUMAN INTELLIGENCE
and control groups. This does not work for
the study of group differences. We cannot
assign people randomly to be young or old,
male or female, Black or White, American
or British. They come as they are. And when
they do, they typically come with many dif¬
ferences besides their group membership.
This makes understanding some differences
extremely difficult.
There are collinearities between group
membership and a host of variables. Con¬
sider the relation between intelligence and
national wealth. (We will go into this in
considerable detail in section 11.5.) There is
a fairly high correlation between national
wealth, measured by gross domestic prod¬
uct per person (GDP/c], and national aver¬
age scores on intelligence tests. There is
also a high correlation between mean intel¬
ligence test scores and indices of physical
health, intelligence, scholastic achievement,
and GDP/c. What are the causal relations?
One could argue that intelligence causes
wealth, or that school achievement causes
wealth, or that wealth makes possible good
schools, which in turn cause intelligence to
rise.
Unfortunately, there have been studies in
which investigators simply ignore the prob¬
lem. They seize upon a single causal variable,
establish that it has a correlation with, say,
intelligence, and spin a plausible tale about
why this is so. Historically, this is the sort
of reasoning that led people to believe that
malaria (Latin, mal aria = bad air] was due
to dank, hot air because there was a corre¬
lation between humid conditions and out¬
breaks of the disease. We now know that
the disease is borne by mosquitoes. Empha¬
sizing isolated correlations might have been
justified in the days before the technologies
for multivariate statistical analyses had been
developed. They are not justifiable today.
Collinearity issues are partly addressed in
quasi-experimental designs, in which we find
two groups that appear to be equal on all the
variables that we believe are relevant, except
one, which may or may not be an exper¬
imental manipulation. This is the sort of
design used when teaching methods are con¬
trasted in comparable school districts, with
one method being used in one district and
another method in another. Similar cases
appear in the literature on intelligence. For
instance, one of the pieces of evidence used
to argue that schooling influences general
reasoning is that in the early twentieth cen¬
tury children's IQ test scores were higher in
isolated rural communities that had schools,
compared to the scores of children who did
not have access to schools.^ The argument
is valid to the extent that the different com¬
munities really were equated on variables,
such as nutrition and health, that could also
influence test scores.
Collinearity issues can be addressed by
means of a statistical technique known as
causal modeling, which is used to deter¬
mine which of several explanations best fits
the data. For instance, the previous chap¬
ter reported an investigation of the relation
between socioeconomic status (SES), intel¬
ligence (as indexed by the SAT] and aca¬
demic achievement.9 Three causal models
were considered: one in which intelligence
and SES had independent influences on aca¬
demic achievement, one in which intelli¬
gence had an influence only as a proxy for
SES, and one in which SES had no direct
influence on academic achievement, beyond
any influence that it had on intelligence. The
third model was shown to be the most accu¬
rate. Note that the question has shifted from
finding a statistically significant relationship
to the more sophisticated issue of selecting
the best model.
Causal modeling is a powerful technique
but is not perfect. When all is said and
done, causal models are based on corre¬
lations, and correlation is still not causa¬
tion. Causal modeling does not establish
what the truth is. The technique gains its
power by rejecting initially plausible mod¬
els for better ones, not by rejecting the null
hypothesis, which almost no one believes in
anyway.
Numerous studies have been done using
what at first appears to be a classic experi¬
mental design. Members of different groups
8 Ceci & Williams, 1997.
9 Sackett et al., 2009.
THE DEMOGRAPHY OF INTELLIGENCE 361
are recruited; within each group participants
are randomly assigned to experimental and
control conditions, and the result of the
experiment-control contrast is compared
across groups. Stereotype threat has been
investigated in this way. People in an
affected group (women, African Americans,
etc.) are either reminded of their group iden¬
tity or not reminded of it, and then given
a test on which members of the group are
said to do poorly. The same logic could
be used to compare test scores obtained by
people who had been given either a newly
invented “IQ pill" or a placebo. In an exten¬
sion of this design, the same experimental-
control contrast is applied within different
groups, for instance, in an investigation con¬
trasting two teaching methods in groups of
Black and White students.
Studies of this sort are excellent ways of
demonstrating that a particular manipula¬
tion can (or cannot) have an effect on test
scores. They do not show whether the effect
actually occurs, or how large it is compared
to other effects, in the world outside the
laboratory.
No method for investigating group differ¬
ences in intelligence is perfect. They all have
something to contribute. The strengths and
weaknesses of the various designs have to
be kept in mind as we consider the many
results that have been obtained in the study
of group differences in intelligence.
11.1.4. Statistical and Measurement Issues
In evaluating studies of group differences
two statistical issues are of importance. They
deal with measurements of the size of a
group difference and the appropriateness of
using tests as a way of assessing intelligence
in different groups of people.
ESTABLISHING THE SIZE OF GROUP
DIFFERENCES
In virtually all cases we shall consider, esti¬
mates of intelligence in one group will not be
uniformly higher or lower than differences
in another group. In fact, this is true of most
human traits. What we need is some mea¬
sure of the extent to which two populations
overlap. The difference between means in
deviation units, d, does this.
Consider two arbitrary groups, with
mean scores on some test M1 and M2, and
with within-group standard deviations S1 and
S2. The deviation score, d, is defined as10
The d measure provides a picture of the
extent to which two groups overlap. The
larger the value of d, the less the overlap.
This is illustrated in Figure 11.1, which shows
the overlap between three groups that are
ordered with respect to their means, and
have identical standard deviations within
each group.
The d measure is a measure of the size
of an effect, just as the correlation coeffi¬
cient, r, is a measure of the size of an effect.
Before interpreting a finding that d is not
large enough to be statistically reliable at
the .05 or .01 level, it has to be shown that
the study had adequate statistical power to
detect the smallest difference of interest. All
the concerns about power that applied to r,
and were introduced in the previous chap¬
ter, also apply to d.
THE APPROPRIATENESS OF USING THE
SAME TEST IN DIFFERENT GROUPS
If we want to measure group differences on a
physical variable, such as height or weight,
the principle is simple. We use a physical
device to make a physical measurement. If
we want to compare, say, the heights and
weights of Asians to the heights and weights
of Africans, there is no concern that tape
measures and scales would behave differ¬
ently across groups. The problem is subtler
when it comes to cognitive testing, because
different types of people may systematically
respond to a test in different ways, thus
tapping different underlying psychological
variables.
10 The equation shown applies to a contrast between
two groups of equal size. The case of unequal group
sizes requires a slight complication in equation 11.1,
but poses no new conceptual issue.
362
HUMAN INTELLIGENCE
Figure 11.1. An illustration of the overlap between distributions. All
distributions have an arbitrary standard deviation of 1. Distribution
A has a mean at o on the abscissa, distribution B has a mean at —1,
and distribution C has a mean at +1. The d (deviation) value for
A - B is 1, for A - C is —1, and for B - C is —2. The amount of
overlap decreases as the d value increases.
To illustrate this, imagine that we decide
to develop a direction-taking test to measure
people’s ability to orient in space. The test
might contain items like this:
Go to the flagpole to the north of your
present position.
Then go east to the church.
Turn north at the church and proceed to
the school.
We then see how long it takes people to
get to the school, and how many deviations
they make from the prescribed route.
Remember that we are not interested in
direction-taking per se. We are interested in
performance on the direction- taking test as
a measure of a more general ability to orient
in space.
If we use a contrast between the speeds
with which men and women arrive at the
school, we have implicitly assumed that
men and women will approach the test in
the same way. But there is evidence to the
contrary. Men often approach problems in
direction taking by constructing a mental
map of the territory and the route through
it. Women tend to memorize the directions
and use them to follow a path.11 The test
11 Hunt, 2002.
evaluates a different underlying trait in men
and in women, so comparing scores would
be valid if we were interested in how well
people did in direction finding, but not valid
if we were interested in making an infer¬
ence about men’s and women’s standing
on an underlying latent trait of orienting
ability.
In order to guard against such errors we
have to have some way of showing that a
test measures the same psychological trait
in different groups. If we are dealing with a
single test, and are concerned that different
problems within the test tap different abili¬
ties in various groups, we look for evidence
of differential item functioning.
The idea behind differential item func¬
tioning can be illustrated by another hypo¬
thetical test, a test of baseball knowledge.
The more you know about baseball, the
better able you will be to answer ques¬
tions about baseball. The question “How
many players are there on a team?” can
be answered by anyone with the slightest
knowledge of the game; the question “What
is a squeeze play?” is a bit more challeng¬
ing. Many Americans who have not thought
about baseball since grammar school can
answer the first question but not the second;
a sports reporter would be able to answer
THE DEMOGRAPHY OF INTELLIGENCE
363
both. More generally, we should be able to
order questions in terms of their difficulty.
Now suppose we give this test to two
groups of people, Americans and Aus¬
tralians. Baseball is played in Australia, so
some Australians know more about baseball
than some Americans. The question is “Does
the test measure the same thing, baseball
knowledge, in both countries?” If it does,
then the order of difficulty of items should
be the same in each country, even though
Australians, on the average, might answer
fewer questions correctly. My intuition is
that the order of item difficulty would be
the same.
But let's go back to the direction-finding
test, and this time the contrast is between
men and women. Because the test measures
orienting ability in men and verbal memory
in women, there is no guarantee that the
order of item difficulty would be the same.
Suppose it is not. In that case we say that
the test shows differential item functioning
across groups. When differential item func¬
tioning is found, conclusions about group
differences on a trait said to underlie a test
cannot be drawn from observations of dif¬
ferences in test scores.
The simplest way to test for differential
item functioning is to calculate the rank-
order correlation between item difficulties
within each group. Anything other than
a fairly high value (.8 or above) is evi¬
dence for differential item functioning. If
the items on a test have been selected using
Item Response Theory (IRT; see Chapter 4),
a more sophisticated test is possible. The
item parameters must be the same in both
groups.
COMPARING GROUPS USING TEST
BATTERIES, SUCH AS THE WAIS
Differential item functioning refers to items
within a test. Many studies attempt to draw
conclusions about groups based on scores
on a battery consisting of several tests. This
would be the case, for instance, in a com¬
parison of WAIS scores. In general, the pur¬
pose of the comparison will not be to com¬
pare raw scores, but to compare the groups
on some underlying factor, such as general
intelligence [g)f that is derived from the
scores on the individual tests. In technical
terms, we want to determine group differ¬
ences on a latent trait [g, R in the g-VPR
model, Gf, Gc, etc.) from observations on
the manifest test scores.
An important and deceptively simple-
seeming case of such a comparison is
Spearman's hypothesis , which Arthur Jensen
attributes to the pioneering psychometrician
Charles Spearman.12 The conjecture is that
differences between groups on virtually all
cognitive tests are largely due to differences
in g , general intelligence. For instance, sup¬
pose that we were to compare the grades
of African American and White high school
students in a variety of subjects - mathe¬
matics, language use, history, art, and so on.
According to Spearman’s hypothesis the dif¬
ferences between groups could largely be
accounted for by differences in underlying
general intelligence.
Spearman’s hypothesis is a statement
about group differences on a latent trait,
g, rather than a difference on observable
test scores. By definition, latent traits can¬
not be observed. So how do we evaluate
the hypothesis? General intelligence, in the
sense of g , is operationally defined as a fac¬
tor underlying scores on several (observable)
tests, such as the sub tests of the WAIS.
Jensen has developed a technique called the
method of correlated vectors as a way to test
Spearman's hypothesis.15 As this method has
been used widely, and erroneously, in the
study of group differences, it is worthwhile
understanding it. This serves as a good intro¬
duction to the more difficult task of under¬
standing a better method.
If Spearman's hypothesis is valid, then
those tests that are good measures of g
should show large between-group differ¬
ences, while tests that are measures of spe¬
cial abilities should show smaller differ¬
ences. The extent to which a test is a
measure of g is determined by its loading
12 Jensen, 1998, p. 372. Jensen credits the hypothesis to
Spearman, but it appears to be Jensen’s summary of
Spearman’s ideas on the topic.
13 Jensen, 1998.
364
HUMAN INTELLIGENCE
on the first (general) factor in a factor anal¬
ysis of a test battery (see Chapter 4). The
extent to which groups differ on each of
the tests within a battery is determined by
the d statistic for comparing groups on that
test. In order to apply the method of cor¬
related vectors we compute the correlation
between the test loadings and the d statistic,
across tests. If the correlation is high, then
the hypothesis is confirmed.
In spite of its popularity, the method
of correlated vectors is seriously flawed.
Depending upon the circumstances, it can
either overstate the size of group differences
on g when they are small14 or, in other cir¬
cumstances, understate them when they are
present?5 This means that we will have to
rethink many of the conclusions drawn by
Jensen and others based upon its use. (It
does not mean that these conclusions are
wrong! It just means that they have to be
held in abeyance until further analyses are
done.) The reasons that the method is wrong
are rather complicated, and so have been
presented in Panel 11.2. Fortunately there
is another method, MultiGroup Confirma¬
tory Factor Analysis (MGCFA), that is much
more satisfactory. Hopefully this method
will receive wider use in the future.
This is not just a statistical nicety. There
are many published conclusions that rest,
somewhat shakily, on results obtained by
using the method of correlated vectors. For
instance, it has been used to make the
very strong statement that African Ameri¬
can and White differences in intelligence are
due to differences along the g dimension.16
The argument was that there is a fairly
high correlation across subtests between
(a) the g loading of the subtest and (b)
the difference between White and African
American scores on the subtest. At first
glance, this is an argument for a difference
between Whites and African Americans on
g. But see Panel 11.2 for the problems of
interpretation.
14 Dolan & Hamaker, 2001; Dolan, Roorda, &
Wicherts, 2004.
15 Ashton & Lee, 2005.
16 Nyborg & Jensen, 2000; Rushton & Jensen, 2005.
11.2. Aging
We all age. Binet assumed, and no one
has disagreed, that cognitive competence
increases with age up to early adulthood.
At the other end of life, things are not so
rosy. The incidence of dementia in peo¬
ple sixty-five and older is 10%, six times
higher than the incidence in the population
at large.17 Dementia does not respect emi¬
nence. Ronald Reagan, the fortieth president
of the United States, died with Alzheimer's
disease at age ninety-three. He is believed
by many to have suffered from early symp¬
toms of the disease during his second term of
office (1985-89), when he was in his late sev¬
enties. However, dementia is certainly not
an inevitable feature of old age. Konrad Ade¬
nauer, the first chancellor of West Germany
following World War II, is largely credited
with guiding his country out of chaos and
into a major place among industrial nations.
He assumed office at seventy-three and left
it at eighty-seven.
Theories of adult lives and aging stress
how people find progressively more con¬
fined niches as they age. This is true in gen¬
eral, but there are exceptions. John Glenn
(1921-) received a commission as a US Marine
Corps aviator at age twenty-two, and served
with distinction as a fighter pilot in both
World War II and the Korean War. Fol¬
lowing the Korean War he became a test
pilot and had a brief interlude as a TV quiz
show contestant. He was selected in the first
cohort of American astronauts, and was the
first American to orbit the Earth. Follow¬
ing his retirement from government service
in 1964 (at age forty-three) he became a
successful businessman in a field that had
nothing to do with aviation. He developed
a strong interest in politics, and served as a
United States senator from 1974 until 1997.
When he was seventy-seven he returned to
space as a crew member on a space shuttle
mission, thus becoming the oldest person to
orbit the Earth.
17 www.freemd.com/senile-dementia/incidence.htm,
downloaded 18 November 2008.
THE DEMOGRAPHY OF INTELLIGENCE
365
Panel 11.2. The Method of
Correlated Vectors and the
Alternative, MGCFA
I first describe in somewhat more detail
than in the main text just what the
method of correlated vectors is. I then
present the problems that it raises.
The first step in applying the method is
to determine that the same relationships
hold between tests within each group to
be compared. This is evaluated by a sta¬
tistical test called congruence that deter¬
mines whether or not the two corre¬
lation (or covariance) matrices can be
assumed to be identical in each group.
IF they appear to be identical, the con¬
gruence condition is said to be satisfied.
In a nonstatistical sense, what this means
is that the observable subtest scores are
related to each other in the same way
in each group. Consider the hypotheti¬
cal direction-taking test described in the
main text. If my intuitions are correct,
this test would be highly correlated with
a test of verbal memory in women, and
with a test of rotational ability in men.
The congruence condition would not be
found.
Jensen assumed that failure of congru¬
ence would be evidence that the tests
within a battery were measuring different
things in each population, and therefore
that Spearman’s hypothesis could not be
tested.
Assuming that congruence is satisfied,
the next step is to factor analyze the com¬
bined data, extract a first factor, which is
the operational definition of g, and corre¬
late the factor loadings of each test with
its associated d statistic, as described in
the text.
A Dutch psychometrician, Conor
Dolan, has pointed out that in order to
test Spearman’s hypothesis the data must
satisfy a property called measurement
invariance. The measurement invariance
property is that the factor loadings for
subtests should be identical within each
group being compared. This ensures that
two individuals, one from each group,
who have identical scores on the latent
traits (which is what we are interested in
but cannot observe) also have identical
scores on the test, which we can observe.
In the examples given in the main text,
measurement invariance would fail for
the direction-finding test if it were to be
incorporated into a test battery given to
both men and women.
Dolan and his colleagues showed that
the method of correlated vectors can pro¬
duce positive findings in situations where
measurement invariance does not hold.*
Using a separate line of argument, Ash¬
ton and Lee have shown that the method
can produce negative findings (the corre¬
lation between loadings and mean differ¬
ences can approach zero) in situations in
which there actually is a difference in g. 1
Dolan advocated using Multi-Group
Confirmatory Factor Analysis (MGCFA)
to evaluate Spearman’s hypothesis and
similar conjectures. The technique is a
variety of confirmatory factor analysis,
the statistical method explained in Chap¬
ter 4, section 4.2. MGCFA is prefer¬
able to the method of correlated vec¬
tors in two ways. One is that it involves
explicit tests for measurement invari¬
ance. The other, which is not negligible,
is that it allows investigators to com¬
pare two hypotheses, rather than eval¬
uating the weaker statement that a sin¬
gle hypothesis is a better than chance
description of the data. The method
of correlated vectors tests Spearman’s
hypothesis against chance, not against
competing hypotheses.
The MGCFA technique has been
known since the early 1980s. It requires
relatively large samples from all groups,
and calls for a good bit of expertise in the
use of the required computer programs.
Therefore, it is often not used when per¬
haps it should be. Hopefully this situa¬
tion will change over time, as the flaws in
the simpler method of correlated vectors
become widely known.
* Dolan & Hamaker, 2001; Dolan, Roorda, &
Wicherts, 2004.
1 Ashton & Lee, 2005.
366
HUMAN INTELLIGENCE
The message is straightforward. There are
changes in cognitive competence over the
adult years, both upward and downward.
They are complex. They are also extremely
important to understand, because age is a
variable that affects every one of us.
11.2.1. Issues
Three different designs have been used in
the study of aging. The simplest is the cross-
sectional design, where the investigator takes
measurements on people in different age
groups, at roughly the same time. The good
aspects of cross-sectional designs are that
they are relatively easy to implement, com¬
pared to other alternatives, and that they
provide a picture of differences in the cog¬
nitive competencies of different age groups
in the population, as it exists at the time of
testing. This information can be projected
forward over short time spans. For instance,
if we know the incidence of senile dementia
in people seventy years and older as of 2010,
and we know the number of people in the
sixty-to -seventy age bracket, it is possible to
estimate the incidence of senile dementia
expected in 2020.
Cross-sectional designs pose problems
of interpretation. It is difficult to obtain
samples in which the participants differ
only on age. For instance, elderly partici¬
pants (seventy and over] will report hav¬
ing had less education, on the average,
than younger adults. This reflects changes in
society, rather than psychological changes
in the individuals, but it produces an
inevitable confound between age and edu¬
cation effects. The general principle is that
in cross-sectional designs age effects are con¬
founded with cohort [Flynn] effects dis¬
cussed in Chapter 10. If we compare, in
2010, twenty-year-olds to seventy-year-olds,
we are also comparing people in the 1990
birth cohort to people in the 1940 cohort.
There is no way to tell whether the differ¬
ence between the twenty- and seventy-year-
olds is an age effect or a cohort effect.
Because cross-sectional studies measure
each individual only once, there is also no
way to distinguish between gradual and
sudden changes in cognitive capability. To
see this, consider two possible physiologi¬
cal events: gradual deterioration of the cen¬
tral nervous system (CNS] and sudden dam¬
age due, for instance, to stroke. Either could
produce slowed decision-making. In a cross
sectional design we would find that, on the
average, decision-making processes slowed
with age. However, because different people
would be evaluated at each age, we would
have no way of knowing how much of the
difference was due to gradual deterioration,
which affected everyone, and how much
was because the older groups would con¬
tain more people who had suffered sudden
CNS damage due to stroke.
In a longitudinal design the same peo¬
ple are studied across several time intervals.
This makes it possible to observe age-related
changes within an individual. The con¬
trast between cross-sectional and longitudi¬
nal results is informative. Intelligence test
scores show considerably less drop over the
adult years in longitudinal studies than they
do in cross-sectional studies. This is at least
in part due to the confounding of age and
cohort effects in a cross-sectional study.
Longitudinal studies are expensive and
time-consuming. They are subject to strong
recruitment effects, for people with low test
scores are more likely to drop out than
people with high test scores. This is par¬
ticularly true if the study involves taking
multiple measures, thus requiring a consid¬
erable investment in time on the part of
the study participants. Nonrandom attrition
effects occur. Unless allowance is made for
this effect, aging may appear to be less debil¬
itating than it actually is. 18
The participants in a longitudinal study
come from just one cohort. Therefore, the
effects of general changes in society are
mixed with the effects of aging. As an exam¬
ple, the percentage of women in Terman’s
study who followed professional careers was
high for its time, but was much lower than
the percentage of women following profes¬
sional careers in the SMPY study, who were
18 See Madhyastha et al., 2009, for an example and
discussion.
THE DEMOGRAPHY OF INTELLIGENCE
367
recruited fifty year later. Was this due to any
psychological difference between women
born around 1910 and those born around
1960, or was it due to the much greater career
opportunities for young women in the 1980s
than in the 1930s?
The gold standard is the cohort-
sequential design, such as the Seattle Lon¬
gitudinal Study described in Panel 9.2, in
which investigators recruit people of differ¬
ent ages at the start, follow them as in a lon¬
gitudinal study, and in addition periodically
recruit new participants and follow them as
well. This allows for a separate evaluation
of cohort and aging effects, and provides
for longitudinal studies of different cohorts.
However, cohort-sequential designs are very
difficult to arrange. Aside from the expense,
the biggest problem is ensuring comparabil¬
ity of the samples developed at each phase
of recruitment. In the Seattle Longitudinal
Study participants were recruited from peo¬
ple enrolled in a health maintenance organi¬
zation (HMO). If enrollments in the HMO
have changed over the now fifty years of the
study, recently recruited participants will
not be comparable in all ways to earlier-
recruited participants. There are also prob¬
lems of selective attrition, as is the case in a
longitudinal design.
These problems make the study of aging
difficult. They do not make it impossible.
The effects of aging will be discussed under
three general headings: studies using psy¬
chometric techniques, studies concentrating
on information-processing and physiological
variables, and what turns out to be a very
important, albeit amorphous, set of findings,
“other cognitive changes.”
11.2.2. A Psychometric View o f Aging
Many psychometric studies of changes in
intelligence with age have utilized Cattell
and Horn’s distinction between fluid and
crystallized (Gf vs. Gc) intelligence. Horn
summarized this work, as of the late twen¬
tieth century, by asserting that Gf peaks in
the mid twenties and then falls fairly rapidly,
with the fall accelerating in old age (sixty-
five plus], while Gc peaks around thirty and
then is maintained at a surprisingly constant
level until old age.19 His conclusion has been
widely accepted. The fact that age has dif¬
ferent influences on fluid and crystallized
intelligence is one of the strongest nonpsy¬
chometric arguments for the importance of
making the Gf-Gc distinction, rather than
collapsing Gc and Gf into a single construct,
general intelligence (g).
In Carroll’s expansion of the Gf-Gc
model, crystallized and fluid intelligence are
two of several broad second-stratum mod¬
els, below g but less specialized than pri¬
mary level abilities, such as vocabulary and
image rotation. Other important second-
stratum abilities are visual and auditory rea¬
soning, short-term memory, and long-term
memory retrieval. In a rather-less-cited part
of his argument, Horn observed that all
the second-stratum abilities except Gc show
declines from roughly age thirty onward.
Horn relied heavily on cross-sectional
studies to draw these conclusions. Such
studies are confounded by cohort effects.
This point is particularly troubling because
the cohort effect is much stronger for mea¬
sures of Gf than for measures of Gc.2° If
Gf has increased across cohorts, a cross-
sectional study will show decrements in
Gf with age. Therefore, the result of an
abbreviated cross-sequential study of the
Woodcock-Johnson testing program is of
considerable interest.
The Woodcock-Johnson tests, mentioned
briefly in Chapter 2, are battery-type tests
generated from the Carroll-Cattell-Horn
three-stratum model of intelligence. The
testing program includes tests suitable for a
wide range of ages, from early childhood to
adulthood. The test batteries contain mark¬
ers for Gf, Gc, and several other second-
stratum abilities defined by the Cattell-Horn
model. They also contain a measure of
“broad cognitive ability,” which is analogous
to the general intelligence [third-stratum]
factor in Carroll’s extension of the model.21
19 Horn, 1985; see also Horn & Noll, 1994, and Cattell,
i987-
20 Flynn, 1987.
21 Carroll, 1993.
368
HUMAN INTELLIGENCE
60
Age at Testing
Figure 11.2. Broad cognitive ability (BCA) factor scores as a
function of age. The lines indicate scores at the two points of time
of testing. Heavy lines show quartiles. From McArdle et al., 2002,
Figure 8, reprinted with permission.
Because the different subtests have been cal¬
ibrated using item response theory (Chap¬
ter 2], they provide a score that can be
treated as a linear scale of the trait under¬
lying each subtest. This makes it possible to
talk about differences in the rate of decline
of various measures of intelligence.22
The Woodcock-Johnson test was revised
and renormed in 1990, to produce the WJ-R
test.23 The renorming was based on an
(approximate] probability sample of the US
population, containing about 6,500 cases.
Subsequently J. J. McArdle, a professor at
the University of Virginia, and his colleagues
contacted and retested just under 1,200 of the
original participants.24 Chronological ages at
the time of first testing ranged from two to
ninety years. The interval between testing
times ranged from one to five years. Fig¬
ure 11.2 presents the raw scores for broad
cognitive ability (gj as a function of age.
There is a peak of intelligence in the early
twenties, and a gradual decline thereafter.
However, there are considerable individ¬
ual differences. Some participants in their
22 Hunt, 2007, Chapter 10.
23 McGrew, Werder, & Woodcock, 1991.
24 McArdle et al., 2002.
seventies have scores that would be consid¬
ered quite high for a twenty-five-year-old.
Different aspects of intelligence peak and
decline at different rates. This is shown in
Figure 11.3, which presents changes in Gf
and Gc as a function of age. The two graphs
clearly follow different trajectories of change
over age. This point is reinforced by some
numbers. Table 11.1 shows the peak age and
the rate of decline (in a transformation of
2 scores, a point that need not concern us
here]. All the second-stratum broad abilities
except Gc behave very much like Gf, crest¬
ing in the mid-twenties and then declining
throughout the adult years. The exception,
Gc, is important because Gc is a better pre¬
dictor of workforce performance than Gf.
McCardle and colleagues’ data clearly
provides support for the three-stratum
model as an appropriate way to understand
the influence of age on intelligence. It is
worth noting, though, that the effects in
their data are not nearly as striking as the
description of age-related declines presented
in some discussions. This can be seen by
contrasting Figures 11.2 and 11.3 with Horn’s
stylized presentation of the data, shown in
Figure 11.4. Gc and Gf do follow different
THE DEMOGRAPHY OF INTELLIGENCE
369
Figure 11.3. Age and change scores for Gf and Gc. Gf peaks sooner than Gc, but falls off more sharply.
From McArdle et al., 2002, Figure 9, reprinted with permission.
courses over the life span, but the difference
does not approach that shown in the stylized
summary.
Schaie has reported data that comple¬
ments the findings of McCardle and col¬
leagues. Figure 11.5, based on data from
the Seattle study, shows the cumulative
age changes, from age twenty-five onward,
for three different factors: verbal meaning,
inductive reasoning, and general intellectual
ability. These factors were chosen because
an examination of the subtests on which
they were based suggests that they should
resemble the Gc, Gf, and “broad cognitive
ability’' (g) scales in the WJ-R battery. Com¬
paring Figure 11.5 to Figures 11.2, 11.3, and 11.4,
we see that the two studies are in agreement
in showing that verbal/Gc measures are sub¬
stantially less resistant to age than measures
of inductive reasoning/Gf.
The studies disagree somewhat about
precisely where the declines come. The
McArdle study indicates declines starting
in the mid-twenties for Gf and in the
mid-thirties for Gc. The Seattle Longitu¬
dinal Study data shows declines starting
Table 11.1. Ages at peak value and rate of decline (in “W units/'
which are linearly related to the IRT logit scale) for various
factors evaluated on the WJ-R test
Rate of Change
Age at Highest per Year,
WJ-R Composite Value Ages 20-75
Fluid reasoning (Gf)
22.8
“•5
Comprehension and knowledge (Gc)
35.6
—.01
Retrieval from long-term memory
18.1
“■4
Short-term memory
24.2
“■3
Processing speed
25.!
-.6
Auditory processing
22.7
“•4
Visual processing
24-5
-•4
Broad quantitative ability
29.0
-•3
Broad academic knowledge
29.8
“•3
Broad cognitive ability (g)
26.2
-•3
Source : Data from McArdle et al., 2002, Table 9, reprinted with
permission.
HUMAN INTELLIGENCE
370
Age, in Years
Figure 11.4. A stylized picture of the age trends for different broad
second-level traits in the Gc-Gf model of intelligence. While the
contrast between the Gf and Gc trends is maintained in Figure 11.3,
the contrast is much less in the data than in the stylized depiction.
From Horn, 1986, p. 52.
considerably later. The truth may lie some¬
where in between. Age trends in the WJ-
R renorming were confounded with cohort
effects, which would lead to an overestima¬
tion of the deleterious effects of aging. In
the Seattle Longitudinal Study participants
have dropped out over time, so there is a
bias toward continued participation by peo¬
ple with higher initial abilities. This would
result in an understatement of the age effect.
There has been continuous debate about
the extent to which the pervasiveness of
the general intelligence factor increases with
advanced age. This is called the dedifferentia¬
tion hypothesis. The hypothesis predicts that
the correlations between measures of differ¬
ent types of intelligence, such as verbal and
spatial-visual reasoning, will increase with
age.
Powerful arguments can be advanced for
the dedifferentiation hypothesis. Declines
in intelligence are associated with declines
in health. Injuries that influence the brain
and nervous system should, on logical
grounds alone, have widespread deleteri¬
ous effects on cognitive performance. Non-
pathological reductions in the prefrontal
cortex and other areas associated with the
working memory-attention-executive func¬
tion complex, which are typical of aging,
are associated with lower fluid intelligence
scores.25 Reduction in the capabilities of the
working memory- attention complex should
have a pervasive effect on almost all cog¬
nitive functions, which should lead to an
increased influence of individual differences
in g upon performance.
Despite this argument, the evidence for
the dedifferentiation hypothesis is mixed.
There is some support for an increase in cor¬
relations between cognitive variables with
age in the Seattle Longitudinal Study.26
However, inspection of the quartile lines in
Figures 11.2 and 11.3 shows that there was lit¬
tle increase in the variance of g with age in
the WJ-R sample. Other detailed multivari¬
ate studies of psychometric cognition and
aging have produced evidence both for and
against the dedifferentiation hypothesis.27
Psychometric studies of aging present a
consistent pattern. Crystallized intelligence
25 See, for instance, Raz et al., 2008. A general review
is provided by Kramer, Fabiani, & Colcomb, 2006.
26 Schaie, 2005, pp. 213-215.
27 de Frias et al., 2007; Juan-Espinosa et al., 2002;
Tucker- D rob & Salthouse, 2008.
THE DEMOGRAPHY OF INTELLIGENCE
371
Figure 11.5. Cumulative age changes from age twenty-five for three
different aspects of intelligence. Scores are shown in T units, where
10 is the variance at age twenty-five. Data from the Seattle
Longitudinal Study (Schaie, 2005), Table 5.1. The “intellectual
ability” scale is not an average of the other two. It includes measures
of spatial reasoning and numerical skill, both of which show
substantial decrements past middle age.
increases to a peak in early middle age,
and remains constant throughout the adult
working years. Fluid intelligence peaks
somewhere in the young adult years and
then, on the average, declines through the
latter half of the working years and old age.
However, some people have high fluid intel¬
ligence test scores well into late adulthood.
11.2.3. Changes in Information-processing
Capacity with Age
The picture with respect to changes in
information processing with age is quite
clear. Cognition gets slower, and the work¬
ing memory-attention-executive function
complex does not function as well as it
used to.
Cognitive slowing can be demonstrated
using either psychometric or laboratory
techniques. In psychometric studies of cog¬
nitive skills the examinee is asked to do
something very simple, quickly, such as
crossing out the "a’s” in a text. Labora¬
tory studies are a bit more complex. Choice
reaction time and perceptual decision tasks
(Chapter 6) are favorites. All methods
demonstrate marked slowing with age. For
example, in the three-stratum model per¬
ceptual speed is a second-level ability, on a
par with Gf and Gc. In the WJ-R study pro¬
cessing speed declined with age more rapidly
than any other second-level ability, includ¬
ing Gf (Table 11.1}. A large-scale survey of
the UK population found marked slowing
of both simple and choice reaction times
with age, with choice reaction time being
the more sensitive variable.28
There is a regular relationship between
the time that young and old individuals take
to accomplish a laboratory task such as a
reaction time task. The time that young
adults take to complete a task increases with
task complexity, and the time that older
people take to complete the same task is a
multiplicative function of the time that the
young adults take. The nature of the rela¬
tionship is shown in Figure 11.6, for over 175
tasks and for multipliers ranging from 1 to 2.
Algebraically, the relationship can be sum¬
marized by
L()ld — &Ly0Ung, (U'Z)
where L is the latency of the task and a is a
multiplier greater than 1.29 The greater the
age disparity, the greater the multiplier.
28 Der & Deary, 2006.
29 Cerella, 1990.
HUMAN INTELLIGENCE
372
Reaction time of younger group (seconds)
Figure 11.6. Reaction times lengthen with age. A
summary of over 175 studies in which older and
younger groups were compared on the same
reaction time task. The time in seconds required
by the older group is shown on the ordinate; the
time required by the younger group is shown on
the abscissa. From Cerella, 1985, Figure 1D,
reprinted with permission.
While the fact of slowing is clear, both
the reason for and the implications of slow¬
ing are less clear. It has been suggested that
the reason for slowing is that as people age
there is a decrease in the reliability of infor¬
mation transmission in the neural system.30
Trying to pinpoint just why this should hap¬
pen would lead us into a discussion of age
effects on the brain, which is not our major
concern. The implications of slowing are of
concern.
Two models have been proposed.31 Cog¬
nitive slowing might cause problems in other
aspects of thought. For example, executive
functioning, allocating attention to concur¬
rent or nearly concurrent tasks (like driv¬
ing a car while listening to the radio], is
possible only if information can be evalu¬
ated as it comes in. A person whose cogni¬
tive processing is slowed is simply less able
to keep up with the world. This can cer¬
tainly be demonstrated in laboratory tasks.
Whether it is a factor in familiar situ¬
ations, in which people have developed
30 For example, see the model proposed by Myerson
et al., 1990.
31 Hartley, 2006; Salthouse, 1996.
strategies for dealing with relevant informa¬
tion processing, is another issue. Adults over
fifty regularly play card games, and often
play them quite well. On the other hand,
they do get into accidents because they do
not react quickly enough to an emerging sit¬
uation.
In addition to whatever direct effects
slowed cognitive processing has, slowing
also serves as a marker for the general state
of the nervous system. Therefore, slowed
processing with age is important both in
itself and as an indicator of general cogni¬
tive health.
Whichever of the two models is correct,
it is clear that slowing is a major but not
the entire cause of declines in intelligence
with age. Figure 11.7 presents an idealization
of results from a Swedish study in which
changes in a general cognitive factor (g) were
determined with and without statistical con¬
trols for changes in perceptual processing
speed. Controlling for processing speed on
perceptual tasks, such as figure identifica¬
tion, virtually eliminates the change in cog¬
nition from age fifty to sixty-five, and slows,
but does not eliminate, the decline past
sixty-five.32
As people age there is a breakdown in
the working memory-control of attention-
executive processing complex. This is shown
by decreased performance on short-term
memory tasks, greater susceptibility to inter¬
ruption, and poorer control on dual tasks,
such as simultaneously monitoring a stream
of visual and auditory signals.33 These behav¬
ioral measures are, in general (a) those that
are most closely related to psychometric
measures of g and Gf and [b] those that are
supported by the frontal cortex-cingulate
cortex-parietal cortex complex.
These results strongly suggest that the
decline in the ability to deal with new prob¬
lems (Gf) is caused by a reduced ability to
construct and manipulate problem repre¬
sentations in working memory. The deficit
is probably due to reduced functioning of
32 Finkel & Pedersen, 2004.
33 Hoyer and Verhaeghen, 2006; Kramer, Fabiani, &
Colcombe, 2006; Salthouse et al., 2003.
THE DEMOGRAPHY OF INTELLIGENCE
373
Initial measure
Controlled for speed
Figure 11.7. Changes in the level of a generalized cognitive trait (g)
over the second half of the life span. The measurement of the latent
trait has been set to a mean of 30 and a standard deviation of 10 to
avoid negative numbers. The change is shown with and without
statistical control for speed on perceptual tasks. Data from Finkel &
Pedersen, 2004, Table 3.
the frontal-parietal system, which shows
anatomical deterioration with age.34 The
failure in functioning appears to be general,
rather than associated with a particular part
of the working memory-control of attention
complex.35
11.2.4. Practical Knowledge, Experience ,
and Wisdom
Given what has been presented so far, the
reader may well have decided that some¬
where around age forty-five people become
slow, easily distracted, poorly focusing dun¬
derheads. But how is it that men and women
in their fifties and sixties occupy most of the
important leadership positions in our soci¬
ety? Why are people advised to go to expe¬
rienced physicians when physicians do not
become experienced until their forties? Why
do we routinely trust our lives to commer¬
cial airline captains, most of whom are in
their fifties and sixties?
Experience counts ... a lot. The typical
test of crystallized intelligence (Gc) evalu¬
ates how well a person knows his or her soci¬
ety. The same Swedish longitudinal study
that documented a drop in general cogni¬
tive functioning from age fifty onward found
increases from age fifty to seventy-five in the
extent to which people could answer general
information questions.36
This finding is typical of many showing
that Gc is either stable or increases until
great age. The same thing is true of more
pointed measures of functioning in life. In
the Seattle Longitudinal Study participants
were asked a number of questions about
what the investigators refer to as “basic living
skills.” These varied from questions about
personal finances to questions about how to
cope with failures in household appliances.
Figure 11.8 shows the results. There was very
little decline in basic living skills until age
sixty-five. This result is typical of several
other studies that have investigated practical
intelligence in the elderly.
Although they may have trouble dealing
with the sorts of novel, out-of-context prob¬
lems that populate IQ tests (and especially
tests of Gf), people in midlife and beyond do
quite well in dealing with realistic decision
making.37 On the other hand, the elderly do
not do particularly well in made-up decision
situations, especially when they are under
time pressure. This could be due in part to
general cognitive slowing, in part to the need
to keep several factors in mind when making
34 Kramer, Fabiani, & Colcombe, 2006; Raz et al., 2008. 36 Finkel & Pedersen, 2004.
35 Salthouse, 1996, 2005. 37 Marsiske & Margrett, 2006; Sternberg, 2003a.
374
HUMAN INTELLIGENCE
Basic Skills
Age
Figure 11.8. Scores on a test of basic living skills, measured on two
different occasions. There is very little decline in basic living skills
until past age sixty-five. Data from Schaie, 1996.
a complex decision, and in part to older indi¬
viduals' tendency to stress accuracy rather
than speed in decision-making situations.
Why is there a difference? Laboratory
decision-making tasks are constructed to
reveal the process of decision making, in
situations where all participants have little
prior experience with the problem at hand.
Outside the laboratory, decision making is
heavily influenced by the possession of rele¬
vant information. 38 People use the heuristics
that are “condemned” in the laboratory, and
they work49 The information-processing
burden shifts from working memory to long¬
term memory and retrieval processes. The
latter processes are relatively immune to
age-related decline until quite late in life.
The heuristics used outside the labora¬
tory rely upon the decision maker's having
experience with similar decision-making sit¬
uations - which is precisely what mature
decision makers have. This sort of reasoning
has been extended to what some have called
“wisdom.” The intuitive idea is that as you
age you acquire wisdom - or at least, you
should. But how do we distinguish wisdom
from practical problem solving?
The theme that I have extracted from
the literature is that practical problem
solving and decision making involve prob¬
lems that the individual faces in the here
and now, while wisdom refers to thinking
about broader, more ephemeral issues, often
involving the course of society. For instance,
a group of German researchers have devel¬
oped the Berlin Wisdom Theory, in which
wisdom is defined as “knowing the rules
and meaning of life.”40 Sternberg has defined
wisdom as knowing how to apply creativity
and intelligence for the common good, by
balancing personal and societal interests.41
Gerontologists claim that wisdom is more
typical of the old than the young. That may
be, but I would like to have clearer defini¬
tions and a better analysis of individual dif¬
ferences within the aged population before
I would unequivocally associate age with
wisdom.
11.2.5. ^ Summary Comment on Aging,
and a Remark on Healthy Aging
The study of changes in intelligence with age
has produced four well-established findings.
They are:
1. Fluid intelligence decreases over the
adult life span, while crystallized intel¬
ligence remains stable or even increases
slightly until people are into the retire¬
ment years.
2. There is a generalized slowing of cogni¬
tion with age. It is measurable as early
as the forties, and can influence men¬
tal competencies outside of the labora¬
tory as we approach the seventies and
beyond. The slowing is pervasive.
3. There is a similar drop in the function¬
ing of the working memory-attention
38 Gigerenzer, 2000; Klein, 1998, 2009.
39 Klein, 2009.
40 Brugman, 2006.
41 Sternberg, 2003a.
THE DEMOGRAPHY OF INTELLIGENCE
375
control-executive functioning system
over the life span. As is the case with
general slowing of reactions, the degree
of deficit accelerates past the mid¬
sixties.
4. Laboratory studies underestimate the
effectiveness of older individuals when
they are faced with familiar problems,
especially in situations in which they
have developed expertise. In such situa¬
tions people in their forties and beyond,
even into their sixties, may actually out¬
perform younger individuals.
It is important not to romanticize the
last finding. It refers to the performance of
individuals who have experienced “healthy
aging.” While, as far as I can find out, this
term has never been defined precisely, it
clearly refers to two things.
First, the individual must not have
experienced serious health problems that
impact on cognition. These include, but
are not limited to, cardiovascular problems
and dementia-producing diseases, such as
Alzheimer’s disease and the later stages
of Parkinson’s disease. The dementias may
unfold gradually, so any population of
apparently healthy elderly people (arbitrar¬
ily, anyone over sixty) will include some pre¬
dementia cases. As the population ages the
dementias will develop to the point that the
individuals cannot live on their own. The
high incidence of dementia in the elderly
poses a major public health problem.
Second, in order to stay healthy the
individual must remain engaged with soci¬
ety. Intelligence, like physical fitness, does
not exist in a vacuum. It requires main¬
tenance. Studies of healthy young adults42
have shown that people vary in intellec¬
tual engagement, and that those who engage
with the world are, in general, more intelli¬
gent than those who do not, where intelli¬
gence is measured by psychometric and lab¬
oratory studies. Whether engagement causes
intelligence, or intelligence causes engage¬
ment, is hard to say. There are probably
reciprocal influences.
42 Ackerman, 1996; Ackerman & Beier, 2001.
Longitudinal studies of the elderly, such
as the Seattle Longitudinal Study and the
Swedish study of aging twins,45 have shown
that this trend is more pronounced as
we grow older. Withdrawal and height¬
ened anxiety are statistically associated with
lower intelligence. This becomes a seri¬
ous problem as people age, because the
frequency of threatening life events will
increase with advancing age. Loss of a job
is far more threatening to a person of fifty
or sixty than to one of twenty or thirty.
Loss of a spouse becomes more likely with
advancing age. How we handle such events
may both tell a good deal about someone’s
present intelligence - in the conceptual
rather than in the narrower psychometric
sense - and have implications for mainte¬
nance of intelligence following the trauma.
11.3. Male-Female Differences
in Intelligence
Are men smarter than women? Or is it the
other way around? It depends on whom you
ask, and how the question is framed.
Historically it has been assumed that
men are generally more competent than
women. This belief has been so widespread
that some powerful women have felt that
they should downplay either their abil¬
ity or their femininity. Panel 11.3 presents
some political examples that have stuck
in my mind. Similar disparities are seen
in other fields of accomplishment. When
Charles Murray compiled a statistical analy¬
sis of major contributions in various fields he
found the ratio of eminent men to eminent
women ranged from 50:1 (mathematics) to
10:1 (literature).44
Historically, there have been entrenched
rules limiting women’s participation out¬
side of home and hearth. In today’s
post-industrial society legal discrimination
against women's participation in various
social roles has been very much reduced.
Informal social rules and attitudes dictating
43 Schaie, 1996, 2005; Wetherell et al., 2002.
44 Murray, 2003.
HUMAN INTELLIGENCE
376
Panel 11.3. Historic Examples and
Historic Commentaries
Here are a few historic examples of
accomplished women who have either
hidden or downplayed their intellectual
characteristics.
Hatshepsut (fifteenth century BCE),
one of the very few women pharaohs, had
herself depicted with a false beard, proba¬
bly to preserve an image of a ruler. Some
two millennia later Elizabeth I of Eng¬
land, one of the most powerful women
who ever lived, famously described her¬
self to her troops as “weak and feeble”:
l know 1 have the body of a weak and
feeble woman . . .
Elizabeth l, speech to English
army assembled to defend against a
Spanish invasion , 1588*
Actually, Elizabeth was not all that
uncertain about herself. It is not clear
whether she actually spoke the words
just quoted, or whether this quotation
was made up by her advisors, in an early
effort at public relations spin. In a better-
documented speech to Parliament she
had said
/ thank God that I am endued with such
qualities that if l were turned out of the
Realm in my petticoat 1 were able to live
in any place in Christendom .
Elizabeth I, speech to
Parliament , 1566*
Given her career, I suspect that the
second quote, rather than the first, bet¬
ter reflected Elizabeth I’s self-image. The
interesting thing is that either she felt it
necessary to assert a “weakness of fem¬
ininity” in the first quote, or, if she did
not actually say this (and the historical
record is weak here), it was necessary for
her supporters, quoting her speech, to say
that she said it.
The image of weak-willed femininity
was still held in England two hundred
years later, when Mary Wollstonecraft
presented the first known argument for
full female participation as movers and
shakers in society. The second chapter of
her book was titled “The Prevailing Opin¬
ion of a Sexual Character Discussed.” It
began
Many ingenious arguments have been
brought fonvard to prove , that the two
sexes, in the acquirement of virtue,
ought to aim at attaining a very dif¬
ferent character: or, to speak explicitly,
women are not allowed to have suffi¬
cient strength of mind to acquire what
really deserves the name of inrtue.
- Mary Wollstonecraft, opening
lines of Chapter 11 of A Vindication
of the Rights of Women . . . (17 92J
Wollstonecraft was certainly right
about what people thought. Five years
earlier the framers of the Constitution of
the United States (1787), considered an
extremely progressive political manifesto
in its time, denied women the right to
vote.
On the other hand, at the time
Wollstonecraft was writing Catherine II
(“the Great,” 1729-96, ruled 1762-96) was
Empress of all the Russias, and an auto¬
crat’s autocrat. She followed Elizabeth of
Russia (1709-62, ruled 1741-62), who was
a powerful figure herself.
Conflicts between women’s actual
accomplishments and widespread beliefs
about the relative abilities of men and
women are not just a modern phe¬
nomenon.
* Shapiro, 2006.
t Ibid.
THE DEMOGRAPHY OF INTELLIGENCE
111
differences in women’s roles vary substan¬
tially across countries and across social
groups within countries. Nevertheless, no
modern society, and probably no past soci¬
ety, assigns exactly the same roles to
men and women. The form of differenti¬
ation varies greatly, ranging from the sub¬
stantial egalitarianism of modern Scandi¬
navia to the much more rigid assignments
of gender rules in conservative Islamic
states.
Although societies vary in what they see
as appropriate roles for men and women,
there is one universal. Childbearing is dic¬
tated by biology. Child care is almost exclu¬
sively a female role, especially for younger
children. This gender-differentiated role
extends beyond caring for one's own chil¬
dren. Nursery, pre-school, and elementary
school teachers are predominantly, although
not exclusively, women.
Outside of the child-care role, assign¬
ments in modern industrial society are more
equitable than they have been in past soci¬
eties, but the extent of the gender differen¬
tiation varies over time and place. Men pre¬
dominate in “traditionally masculine” fields,
such as firefighting, police work, mechan¬
ical trades, military service, and aviation.
There are more women than men in sec¬
retarial trades, nursing, and different vari¬
eties of interpersonal counseling. Some of
these distributions of roles have seen rapid
changes in the past fifty years. Medicine, for
instance, was once almost exclusively a male
occupation. Today women make up slightly
over half the graduating classes from Amer¬
ican medical schools.
The clergy provide an interesting case.
Before 1950 women clergy were virtually
unknown in the Catholic, Protestant, and
Jewish faiths. Today women regularly serve
as clergy in Protestant and Jewish commu¬
nities. The Catholic Church forbids women
to serve as priests. This distinction is inter¬
esting because the duties of the clergy
are entirely intellectual and interpersonal.
Strength, speed afoot, and the ability to ori¬
ent in space, all characteristics that display
strong sex differences, are irrelevant to the
duties of a church leader.
Outside of child rearing, why should sex-
role differentiation be so pervasive? Sto¬
ries (with only indirect substantiation) about
“man the hunter and woman the gleaner"
present a plausible case for evolutionary dif¬
ferences in sex roles, based on men’s supe¬
rior size and strength. But why should these
practices persist in modern society, where
computer programming is far more impor¬
tant than hunting skill? Could it be that
the differentiation is possibly due to male-
female differences in cognition? Or are they
due to social inertia, because we are still
stuck with Neolithic thinking about what
women can do?
These questions do not have easy
answers. Diane Halpern, a professor at the
Claremont Graduate Schools and a well-
respected reviewer of the modern literature
in the field, had this to say:
It seemed like a simple task when I started
writing this book. . . . At the time it seemed
to me that any between-sex differences
in thinking abilities were due to social¬
ization practices, artifacts and mistakes
in the research, and bias and prejudice .
After reviewing a pile of journal articles
that stood several feet high, and numer¬
ous books and chapters that dwarfed the
stack of the journal articles, 1 changed my
mind. The task I had undertaken certainly
wasn't simple and the conclusions that I
had expected to make had to be revised.
Halpern, 1986, Introduction
The next few sections show why Halpern
found the topic so complex.
11.3.1. General Intelligence I: The Evidence
from Studies of Battery -type Tests
The case for differences between men and
women in general intelligence, or g, rests
on three pieces of evidence: the results
from overall scores derived from battery-
type tests, such as the IQ score derived from
the Wechsler tests; the results from factorial
studies in which individual scores are com¬
puted for the g factor derived from a vari¬
ety of test batteries, including both avowed
intelligence tests and batteries constructed
for research purposes; and the results from
HUMAN INTELLIGENCE
378
studies of individual tests, such as the Raven
Matrices tests, that purport to be measures
of g. We will look first at the results from
the battery- type tests, and then examine
the results from the individual tests and
some issues involving tests used for person¬
nel screening.
Analyses of the adult standardization
samples of the WAIS-III and WAIS-R gener¬
ally show a small difference in IQ in favor of
men. The results are consistent across coun¬
tries, running from two to three IQ points
in the United States and Canada45 (in devi¬
ation units, d = .19] to four points ( d = .27)
in China and Japan.46 These results are also
close to the results obtained in earlier stud¬
ies, showing consistency in time.47 There is
a somewhat similar picture when we look
at children’s data. IQ differences are on the
order of one to two points in favor of boys
in both the US and the Netherlands.48
These results are not exclusive to the
Wechsler tests. Deary and his colleagues
report an elegant example, involving data
from the NLSY79, described in panel 9-9-49
The data is of interest because the motiva¬
tion for constructing the ASVAB, predic¬
tion of performance in the armed services,
is rather different than the motivation for
constructing the Wechsler tests, which were
designed for use by clinical and educational
psychologists.
Deary and colleagues contrasted the
scores of brothers and sisters of the sev¬
enteen to twenty-three-year-olds who had
taken the test, thus controlling for family
background. The male-female difference in
deviation scores on the AFQT, the gen¬
eral score derived from the ASVAB and a
good measure of Gc, was —.02, a small, and
not statistically reliable, finding in favor of
females.
For the final piece of data, we look at
a previously referenced, very large survey
of over 70,000 schoolchildren in the United
45 Longman, Sakofske, & Fung, 2007.
46 Dai et al., 1991; Hattori & Lynn, 1997.
47 Matarazzo et al., 1986, Snow & Weinstock, 1990.
48 Born & Lynn, 1994.
49 Deary, Irwing, et al., 2007.
Kingdom, who took the Cognitive Abilities
Test battery at age eleven.50 Two factors
were derived from this test, a general intelli¬
gence factor and a residual verbal factor. No
male-female differences were found on the
general factor. There was a slight advantage
for girls on the residual verbal factor.
What this review shows is either no dif¬
ference or a very small difference in gen¬
eral intelligence in favor of males. Richard
Lynn has argued that there is actually a
greater male advantage in intelligence than
the tests reveal.51 His argument is based on
two claims.
Lynn’s first argument is that girls mature
more rapidly than boys, and that cogni¬
tive competence increases with physiolog¬
ical age, rather than with calendar age. The
male-female difference might be small, and
even negative (reflecting a female advan-
tage) prior to puberty, but a male advantage
would appear after adolescence and con¬
tinue throughout adulthood.
Lynn is correct that male-female differ¬
ences are smaller in childhood than in adult¬
hood, although not very much smaller. It is
not clear whether this should be regarded
as an artifact or simply an observation. By
analogy, male-female discrepancies in height
are smaller (and can show a female advan¬
tage] in childhood and not in adulthood, but
this is not an artifact of the way we mea¬
sure height; girls do get closer to their adult
height than boys do in their pre-teen and
early teen years. To the extent that cogni¬
tive growth mirrors physical growth, one
could argue that girls are, in fact, smarter
than boys during the elementary and mid¬
dle school years, a point that should be
taken into account in situations where entry
into higher levels of education is deter¬
mined by performance in the elementary
years.
Lynn’s second argument is that at all ages
the tests are biased against men. He states,
The adult male advantage of around 4
IQ points obtained by averaging the ver¬
bal comprehension, reasoning and spatial
50 Deary, Strand, Smith, & Fernandez, 2007.
51 Lynn, 1999.
THE DEMOGRAPHY OF INTELLIGENCE
379
Table 11.2. Selected scores and male-female comparison for
brother-sister pairs in the NLSY79 data set
Effect Size
in d Units
Test
Male Mean
Female Mean
(Male-Female)
Word knowledge
22.3
22.9
—.07
Paragraph comprehension
9.2
10.0
—.21
Arithmetic reasoning
15-9
H-7
+.17
Mathematics knowledge
11.9
11.9
0.00
AFQT standardized score
-.034
+ -°34
—.02
AFQT g score
15.08
14.68
.06
Source: Data extracted from Deary et al.# 200 7, Table 1, with permission
from Elsevier.
abilities is not generally found in the full
scale IQ of the Wechsler tests or in the over¬
all IQ of similar tests because the spatial
abilities are typically under-represented in
these tests .
Lynn , 1999, p. 2.
To what extent is this argument plau¬
sible? Summary scores, such as the widely
used WAIS Full Scale, Verbal, and Perfor¬
mance (FSIQ, VIQ, PIQ) scores are deter¬
mined by a weighted combination of scores
on subtests. If men have higher scores on
some subtests, and women on other sub¬
tests, then depending upon the weights
assigned to each subtest you could produce
a summary score that favored men over
women or vice versa, simply by manipu¬
lating the weights assigned to the sub tests.
And it is certainly true that if a test bat¬
tery omits an important ability on which
there are male-female differences, then the
balance of men and women’s scores in an
overall index will be different than it would
have been had the omitted ability been
evaluated.
Arthur Jensen has argued that the way
out of this dilemma is to compare men’s and
women's measurements (factor scores) on g,
defined as the primary factor extracted from
a battery of tests of different aspects of intel¬
ligence. The argument is that the weighting
of individual subtests will then be done by
rational analysis of the data, rather than by
using weights that were arbitrarily assigned
to the sub tests. 52
The two approaches can lead to differ¬
ences in the calculation of overall male-
female differences. This is illustrated by
Deary and colleagues' analysis of the
NLSY79 data. Table 11.2 presents their data
separately for male-female differences on
the four subscales of the ASVAB used to
compute the AFQT. As was mentioned ear¬
lier, the AFQT score showed essentially
no difference between men and women.55
However, when Deary and his colleagues
applied Jensen’s techniques, and computed
male-female differences on a g index derived
from factor analysis, there was a male advan¬
tage of .06 standard deviation units - not a
lot, but still a reversal of the direction com¬
puted from the composite AFQT score.
Several researchers have conducted simi¬
lar analyses of the WAIS and similar batter¬
ies. Some have found evidence for a small
advantage in g for men. Most of these have
relied on the method of correlated vectors,
which, as was explained in section 11.1, is
a questionable technique. Studies using the
better-justified MGFCA procedure find no
52 See Chapter 4 for a detailed discussion of the logic
of this approach.
53 The effect of composite scores upon an overall
index, and upon the deviation scores of the index,
is not determined solely by the weights of each
component. The variance of the components and
the correlation between the components are also
important.
380
HUMAN INTELLIGENCE
difference,54 but these studies can be criti¬
cized for their low statistical power!55
If there are any male-female differences
in general intelligence indices derived from
the commonly used battery-type tests, those
differences are quite small.
No statistical analysis can address a
stronger form of Lynn's argument, that
appropriate subtests are not included at all.
Women generally do better than men on
verbal tests, and men may do markedly bet¬
ter than women on visual-spatial tests, espe¬
cially those tests evaluating the R (rota¬
tional] aspect of visual-spatial reasoning in
the g-VPR model. But does this mean that
those batteries that are now widely used,
such as the WAIS, “underrepresent” visual-
spatial reasoning, as the quotation from
Lynn implies?
To answer this question one would have
to know what the proper representation of
verbal, visual-spatial, and other traits is.
If the test battery as a whole is to be vali¬
dated by its ability to predict performance in
an applied setting, then the appropriateness
of adding or subtracting a particular subtest
can be determined by seeing if the addition
improves accuracy of prediction. A power¬
ful case can be made that adding spatial-
visual tests might do this in some situations,
but might not in others.56 For that matter,
there are some situations in which an argu¬
ment can be made for ignoring spatial-visual
54 See Colom et al., 2000, and Dolan et al., 2006, for
Spain; Jensen, 1998, Table 13.1, for American and
United Kingdom data; and Van der Sluis, Derom,
et al., 2008, for data from Belgium and the Nether¬
lands. Nyborg [2005J reports a substantial male
advantage, but he used a small sample of people
who agreed to participate in extended testing, and
hence may not have satisfied the requirement of
having samples that were equally representative of
men and women, due to recruitment effects.
55 Molenaar, Dolan, & Wicherts, 2009.
56 See Humphreys and Lubinski, 1996, for some
of these arguments. The appropriate subtests to
include in a composite battery depend upon what
the battery is meant to do. The ASVAB legitimately
has different subtests than the WAIS. If a test bat¬
tery is to be used for personnel selection, substan¬
tial legal and ethical issues arise when changing a
test battery results in differential changes in the fre¬
quency of prediction of success for men and women.
It must be shown that these changes increase pre¬
dictive validity.
ability tests in favor of a more extended eval¬
uation of verbal testing.
If the test is intended to measure intelli¬
gence, in the abstract, questions about what
to include in a test battery are unanswer¬
able without a theory of what intelligence is,
defined independently of the tests. This illus¬
trates (once again] the intellectual poverty
of de facto acceptance of the argument that
“intelligence is what the intelligence test
tests.”
11.5.2. General Intelligence II: The
Evidence from Tests Said to Be Markers
for General Intelligence
In theory, a way to study male-female dif¬
ferences while avoiding the problem of hav¬
ing to justify the composition of a test bat¬
tery would be to look at men’s and women’s
scores on a pure measure of g, and compare
the scores obtained in an accurate sample
of a large population, such as the popula¬
tion of a country, where the possibility of
differential recruitment of men and women
into the population would not be at issue.
In practice, it is difficult, if not impossible,
to find such a study. No pure measure of
g exists; the best we have are progressive
matrix tests. The most widely used of these,
the Raven tests, do measure g, but in most
populations they contain a significant visual-
spatial reasoning component.57
Lynn has put the problem succinctly:
Few people will be persuaded that general
intelligence can be so narrowly defined as
to consist solely of fluid ability measured
by the Progressive Matrices. General intel¬
ligence is generally regarded as consisting of
a broader range of cognitive abilities which
would include the verbal and spatial sec¬
ond order factors.
- Lynn , 1999, p. 6
This poses a problem when RPM scores
are used to determine male-female differ¬
ences in g. As will be documented later,
there is substantial evidence that there are
moderate to large male-female differences in
57 Johnson & Bouchard, 2005a, Table 2; Palmer et al.,
i985.
THE DEMOGRAPHY OF INTELLIGENCE
38i
some types of visual-spatial reasoning. Any
difference in RPM scores between men and
women will reflect a difference both in g
and in visual-spatial reasoning. Even if g is
the predominant contributor to the RPM
score, and there are no male-female differ¬
ences in g, a moderate male-female differ¬
ence in visual-spatial reasoning would pro¬
duce a small difference in RPM scores.
The second problem is that the desired
conclusion is a statement about differences
in g, or lack of them, between men and
women in general. Such a statement can be
justified only if it is based on studies where
the participants can be thought of as reason¬
ably close to a probability sample of some
defined large population.
A review of studies conducted prior to
1980 concluded that there were no male-
female differences in RPM scores.58 Lynn
and Irwing properly criticized this review
for having included a large number of con¬
venience samples that made no claim of
being representative of any national pop¬
ulation. They then conducted two meta¬
analyses of their own, one based on the
RPM and another, intended to represent col¬
lege students, that contained studies using
both the RPM and the more difficult Raven
Advanced Progressive Matrices (RAPM).59
They concluded that in adults men score
higher than women by approximately .3
deviation units, which they regarded as
equivalent to 4.5 to 5 points on the IQ scale.
The male advantage was somewhat smaller
in the early teen years, which would be
in accord with Lynn's suggestion that prior
to adolescence girls are both physically and
cognitively more mature than boys of the
same age.
Their conclusions must be taken with
a grain of salt. Lynn and Irwing present
their results as generalizations about male-
female differences in RPM scores across
countries. This means that the representa¬
tion of male and female examinees in a study
must be reasonably representative of males
and females in the relevant country. At a
58 Court, 1983.
59 Lynn & Irwing, 2004a, 2005.
minimum, the male:female ratio in the sam¬
ple should be roughly equal to the likely
male:female ratio in the population. Fail¬
ure to meet this requirement is evidence
that some unknown recruitment effect may
be distorting the results. An unfortunate
number of the studies upon which Lynn
and Irwing base their case do not meet this
criterion.
Here are a few examples. A study said to
represent Israel was actually a study of chil¬
dren in a kibbutz, surely a nonrepresentative
sample of modern Israel. A study involving
200 people was offered as representative of
India, a nation of over one billion. A sam¬
ple of Brazilians aged twenty to forty con¬
tained over 1,900 men and 740 women. This
is a huge distortion of the male:female ratio
for that age group. In another study a sam¬
ple of “American college students” was actu¬
ally taken from a single university, which
stresses its preeminence in engineering and
agriculture.60 The male: female ratio in the
study was approximately 9:2. According to
records I obtained from the university's web
site the enrollments of men and women
were approximately equal during the time
of the study. Such obvious deviations from
representativeness make the application
of meta-analytic techniques questionable.
Additional criticisms have been made of the
college student analysis, on other grounds.61
A somewhat different picture emerges
from the rather scanty reports that have
been made of standardization studies. In
discussing the 1979 standardization of the
RSPM, which analyzed data from a city that
had a demographic profile resembling the
national profile, rather than from a popula¬
tion sample, John Raven reported that there
were no male-female differences in progres¬
sive matrix scores.62 No mention is made of
male-female differences in his discussion of
British and American standardizations in the
1990s, but it is not clear whether a test for
such differences was ever done. A collection
of papers has been published that includes
60 Lynn & Irwing, 2004b.
61 Blinkhorn, 2005.
62 Raven, 2000.
382
HUMAN INTELLIGENCE
reports of several national standardizations
of the Raven tests, and a number of smaller
studies, mostly of children and adolescents.
No male-female differences are reported.63
A major point in Lynn’s argument is that
the difference in RPM scores shifts toward a
male superiority from childhood to adoles¬
cence. Statistically, this would amount to an
age x sex interaction. In seven of the eight
studies of children and adolescents in which
a comparison between the age nine to ten
and age fifteen to sixteen scores could be
made, there was a shift toward better male
performance with increasing age.64
Two conclusions can be drawn from these
frustratingly incomplete results. The first is
that if there is any systematic difference
between men's and women's scores on the
Raven tests, it is a small one. Otherwise
it would show up much more clearly in
studies that do approximate national sam¬
ples. The second is that the difference, if it
does exist, could be due to either the g or
the visual-spatial latent traits that underlie
performance on progressive matrix items.
Carlson and I were right when we stated
our tenth principle for interpreting studies
of group differences: investigators should be
more willing than they are to say “We don't
know.”
Suppose we were to select a randomly
chosen man or woman and try to guess his
or her score on an intelligence test. Our best
guess is the expected value, represented by
E[xm] or E(xw]) depending on whether we
are talking about a man or a woman. Assum¬
ing a normal distribution for intelligence,
this is the mean for the appropriate distribu-
tion; for men, £(%„) = where 11 (xm)
represents the mean of the distribution of
male scores. A similar expression applies for
women, using the w subscript instead of m.
However, we can also expect to miss by
some amount, and the size of the misses will
increase with the variability of the scores.
Let E(xm~ be the expected miss
for men. (The same argument would fol¬
low for women, with w substituted for m.)
The expected miss will be zero, because
overestimates will balance out underesti¬
mates. However, the square of the misses,
[Xyn — /u(%m}]2, will always be non-negative
(and will be positive unless all scores are
equal to the mean score). The more widely
the scores are distributed about the mean,
the larger the value of the expected squared
miss: £((%,„ — //(%m))2). This is the concep¬
tual meaning of variance.
The relative variability in male and
female distributions of scores is measured
by the variance ratio , VR,
11.3.3. The Variance Issue
If there is such a small difference in gen¬
eral intelligence (if any) between men and
women, why do we care? What is all the fuss
about?
Men’s scores on measures of general intel¬
ligence are more variable than women’s
scores. This difference, combined with a
small difference in mean scores, implies that
there will be substantial differences between
men and women at both extremes of the
intelligence distribution, those of above¬
normal and below-normal intelligence. To
explain this a brief excursion into the con¬
ceptual meaning of variance is needed.
63 Raven & Raven, 2008.
64 Lynn & Irwing, 2004a, Table 1.
VR — ^fom))2)
e((xw - n(Xw)yy
VR will be one if men’s scores and women’s
scores are equally variable, greater than
one if men’s scores are more variable than
women's, smaller if women’s are more vari¬
able than men's.
In studies where the sample can be
regarded as approximating a probability
sample of the population, the male:female
variance ratio has been found to be greater
than one, both in overall scores (indicators
of g) and in scores on subtests evaluating par¬
ticular aspects of intelligence, such as verbal
comprehension or spatial-visual reasoning.
Here are some typical observations:
THE DEMOGRAPHY OF INTELLIGENCE
383
123456789
Stanine
Boys
Girls
A Normal
Figure 11.9. The distribution of CAT mean scores in a British
sample of over 300,000 eleven-year-old schoolchildren. The normal
distribution is shown as a reference point. Note the excess of boys
at the extreme values (stanines 1-3 and 8-9] and the excess of girls
in the middle stanines [4-6). Data from Strand et al., 2006,
Appendix B.
1. The WAIS-III provides a general men¬
tal ability [GAI) index, defined by mean
scores on the verbal and nonverbal rea¬
soning sub tests. The means for men and
women, respectively, are 100.1 and 97.9,
a difference of .147 standard deviation
units. The standard deviations for men
and women are 15.1 and 14.7, with a
corresponding variance ratio of (15.1 /
14.76)2 = 1.06.65
2. In the NAEP 2004 data for seventeen-
year-olds, the mean scores for math¬
ematics were men -308, women -305.
The variance ratio was 1.23. For reading,
the scores were men - 278, women - 293.
The variance ratio was 1.3 3. 66
3. In the 2001-03 period 320,000 eleven-
to twelve-year-old children enrolled in
United Kingdom schools took the Cog¬
nitive Assessment Test (CAT-III; see
Chapter 2). The mean score across all
subtests was 99.1 for boys and 99.9 for
girls, an effect size of .05 in favor of girls.
The variance ratio was 1.1 3. 67
65 Lange et al., 2006, Table 2.
66 Data downloaded from nces.ed.gov/
nationsreportcard/lttnde/viewresults.asp, Decem¬
ber 2008. The variance ratios were calculated from
the standard errors, on the assumption that there
was an equal number of men and women.
67 Strand, Deary, & Smith, 2006.
Figure 11.9 uses the British study to
illustrate what this difference in variances
implies for the overall distribution of scores.
In this study there was a trivial difference
in means between boys and girls, but a
reasonably large variance ratio. The figure
shows the distribution of boys’ and girls’
scores in stanines ,68 The standard normal
curve (recalibrated to stanines) has been
superimposed on this distribution. The girls’
scores are leptokurtic, that is, somewhat
more bunched-up than would be expected
if they were normally distributed. The boys’
scores are platykurtic, less bunched-up than
would be expected. They are also slightly
skewed to the left. As a result, there will be a
higher percentage of boys than girls at both
the high and low ends of the distribution,
even when, as is the case in the UK data,
there is virtually no difference in means.
68 A stanine score is a linear transformation and quanti-
tization of a standard score. Stanines are defined for
intervals 1-9, with 5 representing the mean, equiv¬
alent to a standard score of zero, and a standard
deviation of 2. For stanines 2-8 the interval covered
by a stanine has the width of one-half a standard
deviation (stanine range of 1). For instance, stanine
5 covers the standard score interval —2.5 < z < .25.
Stanine 6 covers the interval .25 < z < .75, and so
forth. Stanine 9 covers the interval 1.75 < z < + 00,
and stanine 1 covers the interval - 00 <z < —1.75.
384
HUMAN INTELLIGENCE
Panel 11.4. An Illustration of the
Effects of Small Differences in
Means and Variances
The differences in intelligence test scores
between men and women in both means
and variances are small, in the absolute
sense. In combination they imply sub¬
stantial changes in the ratio of men to
women at the extreme high and low
ends of the distribution. This is impor¬
tant, because socioeconomic contribu¬
tions (loosely, “eminence”) may depend
predominantly upon talent at the high
end of the intelligence scale, while socio¬
economic costs are created dispropor¬
tionately by people at the low end of the
distribution.*
The effects can be seen by comparing
three distributions: the standard normal
distribution (z, mean = o, standard devi¬
ation = 1), which will be used to represent
the female score distribution, F, and three
possible male distributions. They are Mi
(mean = .15 d units, standard deviation =
1), M2 (mean = o, standard deviation =
1.05), and M3 (mean — .15 d units, stan¬
dard deviation = 1.05). As a standard devi¬
ation of 1.05 corresponds to a variance
ratio of mo, these values are in the range
of empirically observed values for male-
female mean differences and male: female
variance ratios.
Figure 11.10 (a) shows the three M dis¬
tributions, superimposed on the F dis¬
tribution. There appears to be little dif¬
ference between them. Figure 11.10 (b)
presents the data from a different per¬
spective. It shows the ratio of males to
females at different points on the z scale,
comparing the three M distributions to
the F distribution, and assuming that
there is an equal number of males and
females in the population. This distri¬
bution is approximately correct for ages
twenty to forty. The male:female ratios
are close to one in the center of the z
scale (i.e., the range of “normal” intelli¬
gence) and strikingly larger than one in
both the upper and lower ranges.
* Gelade, 2009; Herrn stein & Murray, 1994;
Murray, 1998, 2003.
This example illustrates a general prin¬
ciple. Slight differences in the means and
variances of a distribution have little effect
on the distribution of most of the scores, but
can have substantial effects on the upper and
lower tails of the distribution. Panel 11.4 and
Figure 11.10 illustrate how large the effect can
be. The combination of a slight difference
in means and variances has very little effect
on the distribution of intelligence test scores
across men and women in the “generally nor¬
mal” range, say from IQ equivalents of 80 to
120, which is where 80% of all scores lie, but
can produce substantial differences in the
frequencies of men and women among the
top and bottom 10%.
The different frequencies of males and
females at the extreme ends of the distri¬
butions have consequences for educational
programs. Boys markedly outnumber girls
in special education programs, and in other
indices of low but not abnormal intelligence,
by a ratio of about 2:1. 69 At the other end
of the academic spectrum, unless there is
an administrative decision to require equal
numbers of boys and girls, boys usually out¬
number girls in programs for gifted students.
This was true in the Study of Mathemati¬
cally Precocious Youth (SMPY), which was
discussed in Chapter 10. Recall that part of
this study involved three different cohorts,
which can be described as being in the top
1 in 100, 1 in 200, and 1 in 10,000 in the
69 The ratio depends upon the criterion used for
admission to the program. In the United States
those states that offer fewer services, and hence have
a stricter criterion for admission to the program,
have higher M:F ratios (Coutinho & Oswald, 2005).
Similar ratios have been reported for Europe (Skar-
brevik, 2002).
THE DEMOGRAPHY OF INTELLIGENCE
385
Intelligence in d units
Figure n.10. The effect of small differences in means and variances
upon the distribution of scores at various points on the d scale.
distribution of SAT scores. The corre¬
sponding malerfemale ratios were 1.5:1, 2.1:1,
and 11.2:1.
While differences in variance may be part
of the explanation for the excess of men over
women in the extremes of the intelligence
distribution, this cannot be the whole story.
On the positive end of the scale, a d scale
value of 4 (IQ = 160) corresponds to the
“1 in 10,000” cohort in the SMPY. Accord¬
ing to Figure 11.10, the expected male:female
ratio at this point should be 3.5, which is not
even close to the 11:1 observed ratio. The
observed male:female ratio in special educa¬
tion classes (roughly equivalent to d — —2.5,
IQ = 78 and below] is also higher than would
be expected on the basis of differences in
variance alone.
Wendy Johnson and her colleagues at the
University of Edinburgh have suggested a
reason for the overrepresentation of males
at the low end of the distribution.70 They
assumed that the distribution of intelligence
actually consists of two distributions: a dis¬
tribution of the intelligence of normally
developing individuals, which is centered
slightly above the IQ = 100 point, and a dis¬
tribution of individuals who have been sub¬
jected to either biological or environmental
disturbances that disrupt normal develop¬
ment. This distribution, which is consider¬
ably smaller than the first, is centered on the
IQ = 80 point. Assuming that both distri¬
butions have a standard deviation of 15 IQ
points, about 75% of the individuals in the
disrupted group would have IQs above 70,
the usual criterion for the mentally dis¬
abled. Therefore, the disrupted-development
70 Johnson, Carothers, & Deary, 2008, 2009.
386
HUMAN INTELLIGENCE
population would consist largely of people
whose intelligence was in the low normal,
rather than the pathological, range. Johnson
and colleagues further assumed that more
males than females fall into the disrupted-
development population. This is consistent
with a great deal of data showing that males
are generally more at risk for biological dis¬
ruption than females.
Johnson and her colleagues also pointed
out that greater male variability would be
expected if the (yet unidentified) genes
for general intelligence are located on
the X chromosome, because the male
genetic potential would then depend upon a
smaller, and hence more variable, sample of
the alleles for intelligence than would be the
case for women, with two X chromosomes.
The assumption is not unreasonable, for we
know that genes leading to severe cognitive
pathologies are overrepresented on the X
chromosome. A direct test of the hypothe¬
sis will have to wait until the genes for nor¬
mal variations in the genetic potential are
located. Johnson and colleagues' assump¬
tions are sufficient to account for deviations
from the normal distribution in low scores
from two Scottish surveys of intelligence
in eleven-year-olds, taken in 1932 and 1947.
Similar excesses of low scores have been
observed in other data sets.
This leaves us without a proposal to
explain the excess of males at the upper
ends of the intelligence distribution. We
take up this issue subsequently, in discus¬
sions of educational issues, in section 11.3.6.
First we look at male-female differences on
dimensions of intelligence other thang. Here
we find much clearer differences.
11.3.4. The Cognitive Differences
between Men and Women
Although there is at most a small difference
between men and women in general intelli¬
gence, there are substantial differences along
some of the dimensions of intelligence. In
discussing these we will take a top down
approach. We first look at the results from
national surveys. These provide a broad¬
brush view of the nature of sex differences in
cognition, but do not provide details because
the economics and logistics of very large test¬
ing programs rule out close examination of
any one cognitive trait. We then look at psy¬
chometric research studies, which provide a
more detailed look at individual traits, at the
expense of not using nationally representa¬
tive samples, but keeping the constraints of
the conventional testing paradigm. Finally
we look at laboratory studies using the tech¬
niques of cognitive psychology. These stud¬
ies provide a much finer look at individual
behavior, because they relax the constraints
of the testing paradigm, at the cost of
not studying correlations between traits,
and using quite unrepresentative samples.
These approaches complement each other,
in much the same way that the progres¬
sion from public health surveys to laboratory
research provides complementary sources of
information in the biomedical sciences.
We begin with a study of the way in
which scores are distributed in the WAIS-
III standardization sample. Most psychome¬
tric models assume that the scores are dis¬
tributed in accordance with a multivariate
normal distribution, but that is not nec¬
essarily the case. The profiles defined by
the WAIS subtests could fall into patterns
of similar scores, rather than being dis¬
tributed smoothly across the mathematical
space defined by possible test scores. For a
geographic analogy, people's residences in
the US are not characterized by a smooth
distribution across the country; homes are
clustered into cities and towns.
A technique known as cluster analysis has
been used to identify five patterns in the
WAIS standardization.71 The clusters dif¬
fer on two variables: overall performance
level and whether people within the clus¬
ter obtain better scores on perceptual speed
measures than would be expected given
their overall level of performance. Women
do not differ from men in overall perfor¬
mance, but they are overrepresented in clus¬
ters associated with high perceptual speed,
which in effect means rapid recognition of
simple visual figures.
71 Donders, Zhu, & Tulsky, 2001.
THE DEMOGRAPHY OF INTELLIGENCE
387
Table 11.3. Male-female standard deviation unit {d) scores for effect size for different
aspects of intelligence. The NLSY79 survey reported two scores for mathematics -
arithmetical reasoning (use of numerical reasoning in problem solving) and mathematical
knowledge. The same survey also reported two scores for the speed of conducting simple
cognitive operations. One was a simple decoding operation; the other evaluated the
examinee's speed of executing elementary numerical operations.
Survey
Code
Test
Date
N
Reading
Comp.
Mathematics
Nonverbal
Reasoning
Spatial
Ability
Perceptual
Speed
Assoc.
Memory
Project
i960
73;425
~-l5
.12
•°4
•*3
n.a.
“■32
Talent
NLS-72
19J2
16,860
-.05
.24
—.22
n.a.
-•23
—.26
HS&B
1980
25,069
.002
.22
n.a.
•25
— .21
-.18
NLSY79
1980
-.18
.26, .08
n.a
n.a.
-•43; — *23
n.a.
NELS:88
1992
24;599
-.09
■°3
n.a.
n.a.
n.a.
n.a.
Source : Data is from Hedges and Nowell, 1995, Tables 1 and 2.
A similar pattern has been found in
studies of nationally representative ado¬
lescent populations. Table 11.3 shows the
results from four national surveys of peo¬
ple who were tested when they were in
high school, and who have since been fol¬
lowed through their early adult careers.72
While there is some discrepancy between
the results, which is probably due to dif¬
ferences in content between different tests,
two trends stand out. Women do better
than men in tests of reading comprehen¬
sion, speed of simple perceptual operations,
and tests of associative memory, in which
examinees have to recall arbitrary associa¬
tions, such as associating a picture and a
number. Men do better than women on tests
of visual-spatial reasoning and mathematics.
The results from the WAIS III indicate
that in the adult years the female advan¬
tage in verbal operations disappears, but
the advantage in tasks involving rapid, sim¬
ple perceptual recognition and execution
of simple cognitive operations (processing
speed) remains.
These conclusions refer to very broadly
defined abilities. Two psychometric research
studies provide further detail, and, in the
case of the second study, establish a theo¬
retical framework for thinking about these
results.
The Differential Aptitude Battery (DAT)
is a battery of tests developed by the
Educational Testing Service for research
purposes. Figure 11.11 presents male-female
differences on some of the subtests, calcu¬
lated for four standardizations in the United
States, spanning just over forty-three years.
Results quite similar to those shown in Fig¬
ure 11.11 have been obtained in Spain and in
the United Kingdom.75
The consistency of the results is impres¬
sive. Women do somewhat better on tests
of language skills (as opposed to reason¬
ing about verbally presented material) and
on tests of speed and accuracy of simple
operations. Men do markedly better on tests
involving the manipulation of visual images.
Men also do slightly better on tests of verbal
and abstract reasoning.
Johnson and Bouchard have placed
results similar to these in the context
of the g-VPR model.74 In previous work
(reviewed in Chapter 4, section 5) John¬
son and Bouchard had shown that the g-
VPR model provides a good fit to the MIS-
TRA data (Chapter 8). They then removed
the variance in scores associated with the g
dimension, and analyzed the residual scores
on the various tests. The residuals can be
thought of as being the variation in test
72 Hedges & Nowell, 1995.
73 Colom & Lynn, 2004; Strand, Deary, & Smith, 2006.
74 Johnson & Bouchard, 2007a, b.
388
HUMAN INTELLIGENCE
Figure 11.11. Male-female differences in d unit for selected subtests
of the Differential Aptitude Battery, shown over standardizations
from 1947 to 1990.
scores that cannot be ascribed to variations in
general intelligence. The residual variation
could be characterized by three orthogonal
dimensions. The smallest of these, in terms
of variance accounted for, was a “memory
for content of passages” factor. This will not
be discussed further. The other two factors
are more interesting.
Figure 11.12 is a graphic model of Johnson
and Bouchard’s results, showing the two fac¬
tors and their relation to the general intelli¬
gence (g) factor. The two residual factors
are bipolar factors, in the sense that tests
tend to have either high or low loadings on
each of them. Johnson and Bouchard refer
to these two dimensions as verbal-rotational
and focused-diffuse .
The bipolar factors need special interpre¬
tation, for they are not quite the same as
the ability factors identified in the original g-
VPR model. Vocabulary tests and rotational
tests were good markers for the respective
ends of the verbal-rotational factor. What
the result says is that in the MISTRA sam¬
ple (adults!) after general intelligence had been
accounted for people who knew lots of words
tended to be poor at manipulating mental
images, and vice versa. The focused-diffuse
dimension contrasted people who did well
on tasks that require concentration on visual
diagrams with people who did well on tasks
that involve comprehension of verbal argu¬
ments and possession of information about
the world.
Johnson and Bouchard measured male-
female differences along each of the dimen¬
sions. They found only small, nonreliable
differences on the g dimension. There were
large differences [d values larger than .5) on
the subsidiary dimensions. Women tended
strongly toward the verbal ends of both
bipolar dimensions, and tended to have
superior memory for information presented
during the testing session. Men tended to
have higher scores on the “focused” (on
visual objects) and “rotational” ends of the
bipolar dimensions. Once again, though, it
is important to remember that these differ¬
ences refer to performance after removing
variation due to general intelligence.
Johnson and Bouchard have offered an
interesting summary and interpretation of
their results.75 They argue that whenever
a person solves a problem he or she does
so by combining general reasoning ability
(i.e., g ) with the particular mental tools they
have on hand. For instance, many osten¬
sibly visual-spatial problems can also be
solved by verbal reasoning. The results just
cited indicate that men and women dif¬
fer somewhat in the quality of their men¬
tal tools; the verbal tools tend to be better
75 Johnson and Bouchard, 2007b.
THE DEMOGRAPHY OF INTELLIGENCE
389
General intelligence (+)
Rotation: Tasks that
involve manipulation of
mental images.
Tasks that require integrating
information from multiple
sources or analyses, e.g.,
paragraph comprehension,
identifying similarities between
objects.
Figure 11.12. Johnson and Bouchard's model of general intelligence
plus two bipolar factors (verbal- rotational and focus-diffuse). The
bipolar factors represent a trade-off between opposing strengths.
for women, and the mental imaging and
attention-focusing tools tend to be better for
men. Rational application of general intel¬
ligence would thus lead men and women
to adopt somewhat different strategies for
problem solving. Therefore, even though
men and women are essentially equal in gen¬
eral intelligence, they may differ in their
performance on particular tests, because
performance depends upon both general
intelligence (for which there is no sex dif¬
ference) and the residual abilities, where sex
differences may be substantial.
Figure 11.13. Illustrations of two visual-spatial
reasoning tasks on which men tend to
outperform women. See text for details.
Both the national surveys and the psy¬
chometric research studies indicate that the
big differences between men and women are
on perceptual and spatial-visual reasoning
tasks - the P and R dimensions of the g-
VPR model. (One could say much the same
thing about the PS and Vz dimensions of
the three-stratum model.) Laboratory stud¬
ies amplify upon these results. Men tend
to be better than women at tasks involv¬
ing the manipulation of mental images. The
prototypical example is a mental rotation
task, in which two figures must be compared
by moving them about “in the mind's eye.”
Women take longer to do this, on the aver¬
age, and make more errors. The second type
of task on which men's performance is gen¬
erally better than women’s performance is a
field-independence task, where an observer
is required to orient an object to the true ver¬
tical or horizontal, overcoming the effects of
framing figures. Examples of these two types
of task are shown in Figure 11.13. Men also do
better than women on tasks involving judg¬
ment of real or imagined motion.76
76 Hunt et al., 1988; Kimura, 1999, Chapter 5; Law,
Pellegrino, & Hunt, 1993.
390
HUMAN INTELLIGENCE
Figure 11.14. Detecting details in a complex
picture. Count the triangles in the figure above.
This task requires analysis of a static visual
figure. Women tend to outperform men on this
sort of task.
By contrast, there are many spatial-visual
tasks that do not involve any of these skills.
For instance, women tend to outperform
men on tasks that require analysis of static
visual figures. An example, where the task
is to find a component within a picture, is
shown in Figure 11.14.
A possibly related finding is that males do
better than females in way finding, the abil¬
ity to find locations and to maintain aware¬
ness of positions in the environment. This
is true of both children and young adults,
and appears to be related to performance
on spatial orientation tasks, although the
relationship is not a strong one. This result
holds for imagined routes through environ¬
ments, actual orienteering, and acquisition
of knowledge of the environment through
interaction with computer-generated virtual
environments.77
The skills evaluated by these tasks have
applications in everyday life. There are very
large individual differences in our ability to
find our way about the world. And, as would
be expected from Johnson and Bouchard’s
analysis, way finding can be accomplished
by visualizing the surrounding environment
or by using strategies that rely on verbal
memory. Men are more likely to use the first
strategy, women the second.78 There are also
a number of practical situations in which
77 Choi & Silverman, 2003; Malinowski, 2001; Waller,
Knapp, & Hunt, 2001.
78 See Hunt, 2002, for a review of the evidence.
a person must visualize motion or changes
in perspective while viewing a static dis¬
play. These vary from analyzing gear trains
to assembling objects from diagrammatic
instructions - a task that will be familiar and
perhaps frustrating to anyone who has pur¬
chased to-be-assembled furniture kits.
Such observations do not mean that
“women can’t (find their way in the woods,
read maps, be architects, fly helicopters . . . )
as well as men.” The following qualifications
should be kept in mind.
Virtually all tasks we encounter in daily
life admit to multiple solutions. General
intelligence is a far better predictor of
performance than rotational ability, even
though both are important. People can
apply general intelligence to develop a
problem-solving strategy that suits their par¬
ticular strengths.
The claims for male superiority in rota¬
tion and related visual-spatial abilities refer
to statistical trends. Assuming a deviation
difference on a rotation task of .5, we would
still expect 30% of the women to outper¬
form 50% of the men. The differences in the
frequencies of men and women, in favor of
men, would be more extreme at higher lev¬
els of performance.
Visual-spatial abilities, like virtually all
abilities, can be acquired through training.
This is particularly true if the training can
be focused on a narrow set of skills required
for a particular task. On the average, women
do not read maps as well as men.79 Map
reading is taught all the time, to both men
and women, in contexts varying from sports
orienteering to military training. The visual
rotation task (Figure 11.13) was described as
the prototypical task for illustrating male-
female differences in rotation. Similarly, it
can be used to illustrate age differences in
rotation, which can be substantial over the
adult years. People can be trained to per¬
form this task. In one study in my own lab¬
oratory women in their fifties were trained
to the level of performance typical of male
undergraduates.80 These are only two of
79 Boardman, 1990.
80 Berg, Hertzog, & Hunt, 1982.
THE DEMOGRAPHY OF INTELLIGENCE
391
several examples that could be given of the
acquisition of visual-spatial skills through
training.81
Individual differences are generally not
eliminated by training, although they may
be reduced. This is what happened in the
study in my laboratory; both men and
women got better after five days of train¬
ing, but there were still differences between
them in their ability to do the spatial rota¬
tion task. Cognitive traits are not invariant,
in the sense that height is. They are more
like weight: they can be modified, but it can
take a lot of work to do so. The amount of
training required to reach a fixed level of
performance will probably depend upon a
person's initial performance level.
11.3.5. Male-Female Differences in
Cognitive Traits Relevant to Education
This section discusses cognitive differences
between men and women that have a direct
impact upon education. The issue is impor¬
tant, because different educational avenues
lead to markedly different careers in adult¬
hood.
Psychologists and educators think about
individual differences in different ways. Psy¬
chologists are interested in tasks that maxi¬
mize individual differences in human behav¬
ior, try to characterize these differences,
and, especially in recent years, try to relate
behaviorally defined traits to biological sys¬
tems. To illustrate, one of Johnson and
Bouchard’s arguments for preferring the g-
VPR model of intelligence to the Gc-Gf
model is that the behavioral distinctions
in the g-VPR model map onto distinctions
between brain systems, while those in the
Gc-Gf model do not.
By contrast, educators think in terms of
subject matter. They want to talk about cog¬
nitive traits associated with educational con¬
tent. The biggest division of the curricu¬
lum is into topics that are broadly associated
with language arts and topics associated with
mathematics. Accordingly, to an educator
81 See Newcombe, 2007, for a further discussion of this
issue.
the most interesting individual differences
are differences in the ability to deal with
language and mathematics. "General reason¬
ing" is too amorphous a concept, and, to an
educator, perceptual and visual-spatial skills
seem too microscopic.
We need to coordinate these two views,
paying special attention to the ways in which
male-female differences in cognitive experi¬
ences impact on education. A large study
from Germany provides a good place to
begin.82
Martin Brunner and two colleagues at
the Max Planck Institute conducted a psy¬
chometric analysis of the data from a
German testing program involving over
29,000 students, randomly selected from the
seventeen-year-olds in the German school
system. The study was exceptional both for
the representativeness of the sample and
for the care that the investigators paid to
the technical issues concerning group differ¬
ences that were described in the introduc¬
tion to this chapter. They concluded that
the data was best fit by a hierarchical model,
consisting of a general factor (g) and a nested
factor model, in which mathematics or ver¬
bal (reading] abilities contribute to over¬
all mathematics or reading test scores. The
traits of interest to educators, language and
mathematical abilities, appear as specializa¬
tions of a general reasoning factor, which is
not a targeted educational variable.
Brunner and his colleagues then exam¬
ined male-female differences along each
dimension. The seventeen-year-old girls
slightly outperformed the boys on the gen¬
eral reasoning and reading factors [d — -.09
for both comparisons), but the boys
markedly outperformed the girls on the
mathematical factor ( d = .94). This con¬
forms to the popular belief that boys are
better than girls at mathematics, although,
as we will show, this finding ought to be
qualified.
Brunner and his colleagues then con¬
sidered what their findings might imply
for mathematics. They argued that math¬
ematical problems are attacked with a
82 Brunner, Krauss, & Kunter, 2008.
392
HUMAN INTELLIGENCE
-0.3 -0.2 -0.1 0 0.1 0.2 0.3
Figure 11.15. Male-female differences on various educational tests
of reading and mathematics. All studies utilized large probability
samples of the relevant US population. Positive numbers indicate a
male advantage; negative numbers indicate a female advantage.
Data from Hedges and Nowell, 1995, Table 2.
combination of general reasoning skills,
in which men and women are essentially
equivalent, and mathematics-specific skills,
in which men, on the average, exceed
women. As a result, men should do better
than women in those areas of the curriculum
that emphasize mathematics, providing that
the mathematics involved is sufficiently spe¬
cialized to emphasize mathematical rather
than general reasoning skills.
Applying Brunner and colleagues’ rea¬
soning to the typical educational progres¬
sion, what we should see is a progressive
sharpening of differences between men and
women in educational accomplishment as
we move from the general education cur¬
riculum through the undergraduate univer¬
sity years, and then on to specialized educa¬
tion and career achievements in the Science,
Technology, Engineering, and Mathematics
(STEM) fields. That is what happens, but
there are some important qualifications.
WHAT MEN AND WOMEN GET OUT
OF THE K-12 SYSTEM
In the United States public education is
available to everyone from kindergarten
through the twelfth grade. While there is
some variation from state to state, atten¬
dance is usually compulsory through age six¬
teen, and students are very strongly encour¬
aged to complete the entire course, graduat¬
ing at seventeen or eighteen. Similar public
education programs exist in other developed
countries.
Figure 11.15 presents a comparison of the
scores achieved by high school age boys
and girls on the cognitive tests used in sev¬
eral Department of Labor surveys during
the last half of the twentieth century. 83
Throughout this period girls consistently
outscored boys on tests involving language
use, while boys outscored girls by a some¬
what larger margin on tests of mathemati¬
cal skills. Similar results have been obtained
by the National Assessment of Educational
Progress (NAEP). This test has been called
“the nation’s report card” for the evaluation
of language, mathematics, and science skills.
Somewhere between 70,000 and 100,000 stu¬
dents are evaluated each year. In the twelfth
grade girls outscore boys in reading, while
boys outscore girls in mathematics.84
International results are similar. The Pro¬
gram for International Student Assessment
(PISA) is a program in which representative
83 Hedges 81 Nowell, 1995.
84 "Nation’s Report Card,” National Center for Edu¬
cation Statistics.
THE DEMOGRAPHY OF INTELLIGENCE
393
Difference between male and female scores in d units.
Figure 11.16. The difference between male and female scores on
three of the 2003 PISA examinations. Differences are expressed in
deviation units using a nominal standard deviation of 100, which is
the intended standard deviation for the PISA examinations.
schools are selected, and fifteen-year-old
children are evaluated on tests involving
reading, mathematics, science, and prob¬
lem solving. The problem-solving section
presents realistic problems that do not
depend upon specific academic knowledge,
such as finding an efficient route on a bus
line. Figure 11.16 shows the male-female
deviation scores on the PISA reading, math¬
ematics, and problem-solving tests, calcu¬
lated for the United States and six other
economically developed countries. The
consistency in the pattern is striking. Across
all countries males exceed females by a
small amount (d’s consistent, but less than
.1) in mathematics, while females exceed
males by a larger amount (d’s around .3) in
reading. On the problem-solving test, where
an attempt was made to set problems that
did not depend upon reading or mathe¬
matics skills, there was virtually no gender
difference.
Figure 11.16 emphasizes the consistency
of differences between boys and girls, but
does not indicate the fact that the abso¬
lute level of achievement varies a great deal
across countries. This variation is shown in
Figure 11.17. The same pattern is evident in
the science scores - small but consistent
male-female differences and much larger
differences between countries. Fifteen-year-
old girls in Australia, Canada, Germany,
Japan, and Sweden have higher mathemat¬
ics and science scores than fifteen-year-old
boys in Spain and the US.
One more point in the international data
is worth noting. Because males typically dis¬
play greater variance in scores than females,
we would expect the differences between
fifteen-year-old boys and girls to be greater
at high levels of performance than it is on
the average. This is the case. Averaged over
countries, the d for male-female median
mathematics scores is .10. At the ninety-fifth
percentile it is .20. 85
The data for K-12 educational achieve¬
ment presents a remarkably consistent pic¬
ture. At the end of the standard (and often
compulsory) school period mid-teenage girls
have greater language related skills than boys
of their age, while the opposite is true of
skills and knowledge in science and math¬
ematics. In terms of deviation units the
female advantage in language is greater than
the male advantage in science and mathe¬
matics. It is important to remember that this
is a statement about the extent to which the
male and female score distributions differ
with respect to each other. It is not a state¬
ment of differences in absolute skill, because
there is no metric by which we can compare
a difference in mathematical knowledge to
a difference in language skill.
85 Machin & Pekkarinen, 2008.
394
HUMAN INTELLIGENCE
■ Male
Female
■ Male
Female
560
| 540
I 520
1 s 500
<U QJ
-g 2 480
£ 460
2 440
■ Male
Female
Figure 11.17. Mean scores of fifteen-year-old students in selected
countries on the PISA 2003 tests of mathematics, science, and
general problem solving. Data from the National Center for
Educational Statistics, Digest of Educational Statistics, 2007,
Table 389.
COLLEGE AND UNIVERSITY
UNDERGRADUATE EDUCATION
College and university students represent
an important population, because this pop¬
ulation of young adults contains most of
the people who will be leaders of soci¬
ety, both in the dramatic sense of provid¬
ing a few highly visible leaders, and in the
perhaps more important sense of providing
the many business managers, entrepreneurs,
technicians, and professionals who will con¬
stitute the economically and socially most
productive segments of society.86 As is well
known, in the last fifty years there has been
a tremendous expansion of social and eco-
86 Gelade, 2009.
nomic opportunity for women within the
college-educated segments of society. There
is considerable reason to be interested in dif¬
ferences between the cognitive skills of men
and women within this group.
It is important to remember, though, that
results obtained by studies of people who are
either in or about to enter undergraduate
education do not necessarily generalize to
the population at large. Since the early 1980s
more women have enrolled in college than
men. As of 2008, there were approximately
three undergraduate men for every four
women. This situation arises because at a
given level of high school academic achieve¬
ment a woman is more likely to take the
first steps toward undergraduate education
THE DEMOGRAPHY OF INTELLIGENCE
395
Male
S- Female
Figure 11.18. Mean SAT mathematics scores for men and women,
from 1988 to 2008. The d values for these comparisons range from
.33 to .41. Source: College Board Report, College Bound Seniors, 2008.
than a man is.87 Accordingly, female under¬
graduates are a somewhat academically less
select subpopulation of all women than male
undergraduates are of all men. While this is
not quite accurate, you can think of the two
populations as being roughly the top 45% of
men and the top 55% of women.
One implication of this is that male-
female differences will be algebraically
larger in the undergraduate population than
in the high school population. The effect
will be further exacerbated by the vari¬
ance effects described earlier. Men’s scores
will be increased, overall, because of a high
male: female ratio in the population of peo¬
ple who have high test scores. The excess
of men with low test scores will not affect
the undergraduate means, because people
with low test scores are unlikely to enroll in
undergraduate programs.
These trends are seen in the test scores of
entering students. In the high school popu¬
lation, as a whole, the women’s mean read¬
ing score is above the men’s mean by about
.2 standard deviation units, and men exceed
women in mathematics test scores by about
.1 standard deviation units. The 2008 SAT
reading scores were 504 for men and 500
for women, while the mathematics scores
were 533 for men and 504 for women. Using
the nominal 100-point standard deviation for
SAT sections, this translates into a trivial .04
87 Hunt & Madhyastha, 2008.
d male advantage in reading, and a nontriv¬
ial .33 d advantage in the mathematics (now
renamed "reasoning”) portion of the test.
This is the sort of magnification of male-
female differences that would be expected
because of the differential recruitment of
men and women from the high school to
the undergraduate population.
The male-female discrepancy in SAT
mathematics scores is not a recent phe¬
nomenon. Figure 11.18 shows the SAT math¬
ematics (SAT-M) scores of college-bound
seniors from 1988 until 2008. Although the
levels of scores for both men and women
vary from year to year, the difference
between means is remarkably consistent.
In 2005 an approximately equal num¬
ber of men and women earned bachelor’s
degrees in Science, Technology, Engineer¬
ing, and Mathematics (in the jargon of the
US National Science Foundation, the STEM
fields). This was a change from thirty years
earlier, when the ratio of men to women
earning bachelor's degrees in these fields was
approximately 2:1. 88 However, because the
undergraduate male:female ratio is 3:4, the
1:1 ratio in the STEM fields implies that a
man is about 4/3 more likely to major in a
STEM field than a women is. The extent to
which this disparity is due to cognitive com¬
petence, personal interests, or social pres¬
sures is impossible to say.
88 Source: National Science Foundation.
HUMAN INTELLIGENCE
396
Figure 11.19. The relation between SAT-M scores, grades (A, B; C,
D, E), and gender in three levels of college courses. Within course,
men have higher SAT-M scores than women who have comparable
grades. Source: Wainer & Steinberg, 1992.
Women are not entering the STEM fields
and then dropping out because they find
the work too difficult. Given constant SAT-
M scores, women consistently outperform
men in mathematics courses. Hq The differ¬
ence can be striking. Figure 11.19 shows the
results from a very large (N ~ 49,000) study
of college students in a variety of Amer¬
ican universities.90 SAT-M scores increase
as a function of the level of mathematics
involved. This reflects the unsurprising fact:
people with high SAT-M scores are more
willing to enroll in mathematics courses than
people with low scores. Within each type
of course, SAT-M scores increase as the
course grade increases. This shows that the
SAT-M is a valid predictor of accomplish¬
ments in mathematics classes. Both of these
trends hold for men and for women.
But then we come to a paradox. Within
each course and grade level, women receive
higher grades than men, even though they
have lower SAT-M scores. In terms of edu¬
cational outcomes the difference can be
substantial. Women who receive Bs have
SAT-M scores lower than the men who
receive Cs and Ds in the same class. This is
a striking example of a general tendency for
89 Clark & Grandy, 1984; Lynn & Mau, 2001; Ramist,
Lewis, & McCamley-Jenkins, 1994; Wainer & Stein¬
berg, 1992.
90 Wainer & Steinberg, 1992.
SAT scores to underpredict women’s educa¬
tional achievements in the early undergrad¬
uate years.
I have heard three explanations offered to
explain the paradox. One is that the test is
an objective measure of mathematics abil¬
ity, while grades are a subjective measure
based on the instructors’ decisions. There¬
fore, men “really” possess more mathemat¬
ics ability, but women are able to present a
more favorable impression to instructors.
This explanation strikes me as a plausible
argument for low correlations between test
scores and grades in courses where there is
a substantial subjective component to grad¬
ing, such as a course in English literature,
but I do not see how the argument applies to
lower-division college mathematics courses,
where right answers are clearly defined.
A second argument is that women simply
work harder to get grades.
The most probable explanation is that
women's stronger work motivation com¬
pensates for their lower test scores.
- Lynn &£ Irwing, 2004a, p. 495
Presumably what Lynn and Irwing meant
was “lower ability as indexed by test scores”
(which in this case referred to progressive
matrix tests rather than the SAT), rather
than to the test scores themselves, as the
test scores per se are not involved in grade
assignment. The motivational argument is
THE DEMOGRAPHY OF INTELLIGENCE
397
not unreasonable, for women do report
doing more homework than men.91 How¬
ever, this implies a false dichotomy between
intelligence and motivation as evidenced by
behavior as a student. Talents such as good
time management and establishing priori¬
ties among goals are part of intelligence, in
the conceptual sense, as much as the talents
required for taking tests.
A third possibility is that tests of math¬
ematical aptitude, such as the SAT-M, are
influenced by a psychological trait on which
there are sex differences, but this trait either
does not contribute to mathematical perfor¬
mance outside the test, or does so, but has
more influence on test performance than it
does on in-class performance. Visual-spatial
ability, the R dimension of theg-VPR model,
has been suggested as a possibility, both for
the SAT and for progressive matrix tests.92
Alternatively, there might be some ability
that is not evaluated by the test, but that is
important in the study of mathematics and
is possessed by women more than by men.
My comment about time management is an
example of such an explanation.
When all is said and done, though, we just
do not know what the link is between male-
female differences in test performance and
in performance in mathematics. The prob¬
lem becomes more acute as we look at dis¬
crepancies between men and women at a
higher level of analysis, the pursuit of careers
in the STEM fields.
POSTGRADUATE EDUCATION
AND CAREER DEVELOPMENT
It is difficult to say anything succinct
about male-female differences in postgrad¬
uate education and career development in
general, because any statement has to be
qualified by considering the field involved.
Postgraduate education is itself so varied
that general statements about how men and
women progress through curricula as differ¬
ent as mathematics, education, medicine,
and the law are equally suspect. What we
can do, however, is to look at some of the
91 Mau & Lynn, 2000, 2001.
92 Casey et al., 1995; Lynn & Irwing, 2004a.
highly publicized differences in outcomes
between men and women. Socially, what
has been of particular concern is the scarcity
of women in the Science, Technology, Engi¬
neering, and Mathematics (STEM] fields.
The disparity is longstanding. Charles
Murray's statistical survey of five thousand
years of human accomplishment uncov¬
ered very few eminent woman scientists or
mathematicians.93 This is hardly surprising,
owing to restrictions on women’s activities
that were enforced by various human soci¬
eties until modern times, and even now out¬
side the industrially developed countries.
Within these countries lifting of the restric¬
tions is quite recent. In 1925 Cecelia Payne-
Gaposchkin (1900-1979) became the first
woman to receive a doctorate (in Astron¬
omy, a STEM field) from Harvard Univer¬
sity. A few years later, in 1934, Grace Hopper
(1906-1992) became the first woman STEM
graduate (in Mathematics) at Yale Univer¬
sity. (Hopper later developed COBOL, one
of the early computer programming lan¬
guages, and was the first person to use the
term “bug” to describe an error in pro¬
gram execution.) In 2007 Drew Gilpin Faust
became the first woman president of Har¬
vard. Hanna Holburn Gray served as acting
president of Yale for one year in 1977. Things
have changed, but perhaps not at breakneck
speed. Harvard was founded in 1636, Yale
in 1701.
The national picture is similar to that at
Harvard and Yale. Figure 11.20 shows the
percentages of women receiving doctorates
in several relevant fields, contrasting 1977
and 2007. There has clearly been a great
increase in the number of women receiv¬
ing advanced degrees. However, the pat¬
tern for type of degree has changed very
little over a thirty-year period marked by
major advances in opportunities for women.
Today women greatly outnumber the men
who receive doctorates in Education, and
are greatly outnumbered by them in Math¬
ematics, the mathematically oriented phys¬
ical sciences, and Engineering.
93 Murray, 2003.
39»
HUMAN INTELLIGENCE
■ % women 2007
% women 1977
Figure 11.20. The percentages of women among the recipients of
doctoral degrees, 1977 and 2007. Source: National Science
Foundation Survey of Earned Doctorates, 2007.
Why have the higher reaches of the
STEM fields remained so weighted toward
men? In a widely publicized speech Faust's
immediate predecessor at Harvard, the
economist Larry Summers, proposed three
reasons for the discrepancy. Two were
social: the existence of conscious or uncon¬
scious prejudice on the part of hiring and
promotion committees, and a preference
for a personal lifestyle that is not compati¬
ble with the workaholic standards Summers
associated with very high productivity in the
STEM fields. The third was biological. Sum¬
mers correctly observed that there is a strik¬
ing disparity between the numbers of men
and women who have very high scores on
tests of mathematical ability and achieve¬
ment. He then speculated that women's
brains are organized in a way that makes
acquiring the high level of analytic skills
needed in STEM research more difficult for
women than for men.94
Summers’s remarks created such a
firestorm that he subsequently resigned
from the Harvard presidency. The incident
is described in Panel 11.5. Going into all the
possible social and biological reasons why
women might be underrepresented in high-
profile STEM positions is well beyond the
94 See the earlier discussion of disparities between men
and women in the SMPY program. For years men
have outscored women on both the mathematical
and verbal sections of the Graduate Record Exami¬
nation (Grandy, 1999).
scope of this book. The controversy does
provide a good entry point for a more gen¬
eral discussion of why there might be dif¬
ferences in intelligence between men and
women.
11.3.6. The Causes of Cognitive
Differences betiveen Men and Women
Professionals in the field were far more
nuanced in their reaction to Summers’s
remarks than political figures and aca¬
demic leaders. The American Psychological
Association (APA) and the Association for
Psychological Science (APS), the two pro¬
fessional organizations most involved, com¬
missioned reports by people well known for
their research in the field. The APA reports
resulted in a book of contributed chapters,
several of which I have cited.95 The other
report took up an entire issue in one of the
APS 's journals. Here are two sentences from
its concluding paragraph.
There cannot be any single or simple
answer to the many complex questions
about sex differences in math and science.
Early experience, biological constraints,
educational policy and cultural context
each have effects , and these effects add and
interact in complex and sometimes unpre¬
dictable ways.
- Halpem et aL, 200 7
95 Ceci & Williams, 2007.
THE DEMOGRAPHY OF INTELLIGENCE
399
Panel 11.5. The Lawrence
Summers Affair
On January 14, 2005, Lawrence Summers,
at that time president of Harvard Uni¬
versity, gave a speech addressing the fact
that women are underrepresented in the
STEM fields, and even more underrep¬
resented at the top of those professions.
Summers subsequently commented that
he had been asked to be provocative. He
was.
Summers proposed three causes for
the discrepancy: conscious or uncons¬
cious prejudice against women, women’s
distaste for the intense professional com¬
mitments required to rise to the top of
the STEM fields, and women’s difficulty
in acquiring mathematical and scientific
reasoning skills. Summers noted that a
relatively small percentage of women
achieve high scores on tests of mathemat¬
ical reasoning, and suggested that biolog¬
ical differences between men and women
might contribute to the disparity in high-
level mathematics skills.
Several prominent women scientists
left the meeting in protest. The high-
visibility magazine Science published a
letter signed by seventy-three prominent
academics protesting his statements.*
The Harvard faculty voted no-confidence
in Summers, and in 2006 he resigned
his post, returning to his position as
a Professor of Economics. He hardly
retired to pasture. In January of 2009 he
was appointed Chair of the Presidential
Council of Economic Advisors, making
him the chief White House advisor on
economic matters.
Summers's remarks on women
undoubtedly contributed to his more or
less forced resignation, but it is unlikely
that they were the only factors. By all
accounts, Summers was a forceful exec¬
utive in an institution that was accus¬
tomed to a great deal of collegiality and
decentralized decision making. During
his tenure at Harvard he clearly ‘Tubbed
a number of people the wrong way” as
he sought what he perceived as urgently
needed changes in an historic institu¬
tion.* As is often the case, the incentive
to take drastic action stemmed from sev¬
eral causes.
The politics behind Summers’s resig¬
nation are not relevant to our discussion
of intelligence. The letter to Science is,
because it indicates both the passions that
are involved in discussions of group dif¬
ferences in intelligence and the beliefs
held by highly influential people who
have not studied the topic in depth.
Here are two quotes from the letter to
Science:
There is little evidence that those scoring
in the very top of the range in standard
ized tests are likely to have more suc¬
cessful careers in science education. . . .
And
We are concerned by the suggestion that
the status quo for women in science may
be natural , ineintable, and unrelated to
social factors.
Muller et al, 2005, p. 1043
Although the list of seventy-three sign¬
ers of the letter included prominent aca¬
demic scientists and science administra¬
tors, it did not include any of the major
figures who do research on individual dif¬
ferences in cognition. 1 doubt that many
of them would have signed the letter, for
the first statement is demonstrably false.
By 2005 the results of the Study of Math¬
ematically Precocious Youth (i.e., people
whose SAT-M scores were in the top 1%)
were well known to professionals in the
field. These results document the stun¬
ning success of people whose scores were
“in the very top of the range in standard¬
ized tests.” *
What about the second statement?
Summers never said anything about the
status quo being natural, inevitable, or
(continued)
400
HUMAN INTELLIGENCE
Panel 11.5 (continued)
unrelated to social factors. On the con¬
trary, he specifically listed two social
factors that he thought contributed to
the disparity: discrimination and conflicts
between professional and family goals.
The fact that Summers mentioned a pos¬
sible biological explanation for differ¬
ences in men's and women's intelligence
was equated with a denial of social causes.
Evidently the academic leaders who
signed the Muller letter felt that they
knew what the truth is, for they either
felt no need to consult with experts in the
field, who would have been easily avail¬
able to them, or they decided to disregard
far more nuanced expert opinions.
* Muller et al., 2005.
* Bowley, 2006.
* Park, Lubinski, & Benbow, 2006,
Let us look at some arguments for social
and biological causes.
SOCIAL CAUSES
Summers’ thought that one of the reasons
that there are more men than women in
the STEM fields is that work in these fields
is simply more interesting to men than
to women. If Summers were correct, men
should have a higher participation rate than
women in the STEM fields even within a
population of talented men and women,
where ability to enter the field is not an
issue, but interest is. And there is such a
group, the SMPY participants, people who,
in their teens, were in the top 1% of their
cohort in terms of academic ability tests. The
SMPY participants were contacted twenty
years later, when they were in their thirties,
to see where their careers had gone.
There were career differences between
men and women. However, career choices
were determined more by interest than by
gender. It is a rough generalization, but not
too far off the mark, to say that partici¬
pants who had primary interests in things
and abstract ideas tended to follow careers in
mathematics and the sciences. Participants
who had interests in people and social issues
followed careers in the humanitarian/social
issues-oriented professions. Participants also
differed in the extent to which they valued
careers or families. Those participants who
had strong family orientations were under¬
represented in the STEM professions, which
are notoriously demanding of time.
Men tended to fall more into the
“things-ideas-career” pattern, while women
tended toward the “people-family orienta¬
tion” pattern.96 However, examples of each
pattern occurred in both highly talented
men and women. The codirectors of the
SMPY, David Lubinski and Camilla Ben¬
bow, point out that difference between
mens’ and womens’ interests alone would
create disparities in the extent to which men
and women choose to work in the STEM
fields.97
The SMPY participants had acquired the
cognitive talents they needed, but differed in
interests. What about the more general issue
of how men and women come to display
disparities in mathematical aptitudes, and in
language skills and visual-spatial reasoning?
Cognitive skills improve with practice,
a psychological law that Diane Halpern98
said is as certain as the law of gravity. [She
was right.) Sex role differentiation is a fact
in every society, although the restrictions
placed on permissible gender roles have
varied greatly both throughout history and
across societies contemporaneously. There
is considerable evidence that boys and
girls do have different learning experiences
that might impact on their acquisition of
visual-spatial skills, and concomitantly of
96 Benbow et al., 2000; Ferriman, Lubinski, & Benbow,
2009; Lubinski et al., 2006; Wai, Lubinski & Benbow,
2005.
97 See Halpern et al., 2007, pp. 31-39, for a good dis¬
cussion of these social issues.
98 Halpern et al., 2007, p. 4.
THE DEMOGRAPHY OF INTELLIGENCE
401
mathematics. A recent review" docu¬
mented the following statements:
In American society parents are more
likely to provide analytical, causal oriented
explanations to boys than to girls.
Middle school children generally accept
the stereotyped view that females do not do
well in mathematics and spatially oriented
tasks.
Male children begin to play and explore
away from home at younger ages than
girls do.
Mathematics and science teachers tend
to direct their discussions more toward boys
than girls.
Why do these differences arise? Is it
because boys and girls draw forth differ¬
ent behaviors from the adults, or because
the adults initiate these behaviors for either
social or biological reasons? These are dis¬
tal explanations; the proximal facts are
that when people are provided differential
learning experiences they will learn differ¬
ent things. Psychologists cannot conduct an
experiment to prove that this is a cause of
differences in men’s and women’s perfor¬
mance on mathematical tasks - for people
cannot be randomly assigned to lifestyles!
There is an interesting natural observation
that is consistent with the differential learn¬
ing explanation.
In forty of the forty-one countries that
participated in the PISA mathematics assess¬
ment program fifteen-year-old males out¬
performed fifteen-year-old females.100 The
sizes of the differences varied considerably.
Nations also vary in the extent to which
they differentiate between male and female
social roles, including education. These dif¬
ferences are reflected in the World Eco¬
nomic Forum’s Gender Gap Index (GGI),
which reflects the relative economic and
social opportunities offered to men com¬
pared to those offered to women. The GGI
correlates negatively (r = — .55) with the dif¬
ference between men and women’s scores
99 See Hyde, 2007, for detailed references.
100 National Center for Educational Statistics, Digest of
Educational Statistics, 2008, Table 389. The excep¬
tion was Thailand.
on the 2003 PISA tests of mathematical com¬
petencies in fifteen-year-olds. The closer the
economic and social opportunities for men
and women are to being equal, the smaller
the male advantage in mathematics.101
The differential learning explanation for
male-female discrepancies in visual-spatial
reasoning and (perhaps as a derivative) in
mathematics is a compelling one. However,
it fails to explain a sometimes overlooked
fact. Why is it that the discrepancy in mathe¬
matics appears on tests but not in class work?
Two explanations have been offered -
one that blames the tests and the other that,
in a sense, blames society.
The explanation that blames the test itself
has two aspects. One is that the test ques¬
tions tend to be interpreted in different ways
by men and women, and that the differ¬
ence in interpretation hurts women’s scores.
To take an extreme example, suppose that
all the word problems on a mathematics
test used sports examples. If women were
less familiar with sports than men are, they
would be at a disadvantage not because they
had lower mathematical skills but because
they had difficulty mapping from the verbal
statement of the problem to the appropriate
mathematical model.
This argument may have had some valid¬
ity at one time, but is unlikely to be valid
today. All the major tests currently in use
are subjected to elaborate statistical anal¬
yses to ensure that the tests measure the
same underlying latent trait, whatever it is,
in both men and women. This sort of error
is far more likely to occur in a locally gener¬
ated class examination than it is in a national
testing program.
The second reason the tests could be
blamed is that they might fail to measure
(or fail to adequately weight) some trait
that is required for performance in class¬
room mathematics, and that is possessed
to a greater extent by women than by
men. We have already encountered this
101 Guiso et al., 2008. See supporting online material for
details of the computation. Interestingly, there is no
correlation between the GCI and the maleifemale
variance ratios in different populations (Machin &
Pekkarinen, 2008).
402
HUMAN INTELLIGENCE
sort of objection, in discussing Lynn and
Irwing's hypothesis that there is a differ¬
ence between test performance and class¬
room performance because women are more
motivated to do well in class (or study more
intelligently!). This seems to be likely.
The explanation that blames society is
that stereotype threat, as described in the
introductory section of this chapter, may
depress women’s test scores. The argument
is that women think they will do poorly,
and therefore give up more rapidly when
faced with difficult mathematical problems.
There is an elegant demonstration that this
can happen. It involved a study of Asian
women college students in a North Ameri¬
can university.102 The women first answered
questions designed to emphasize either their
Asian identities or their identities as women.
They then attempted to solve mathemat¬
ics problems. The idea was to produce two
contrasting stereotypes - the belief that
Asians do well in mathematics and the
belief that women do poorly. The study
worked; students who were induced to
think of themselves as Asians did better
than students who were induced to think
of themselves as women, even though the
assignment to different groups had been
random.
Experiments like this show that the
stereotype threat effect exists. There are
substantial questions about the size of
the effect in situations in which the test
has real consequences, such as a college
entrance examination. In such situations
examinees’ motivation to do well might
override any anxiety introduced by stereo¬
type threat. The evidence is that it can, for
reminding women of their gender has lit¬
tle effect on their performance on either
high-level or low-level real examinations in
advanced placement and community college
classes. 10? If stereotype threat can be over¬
ridden by motivation in these situations, it
seems logical that it would be overridden
in even higher-stakes situations, such as col¬
lege entrance examinations. Direct evidence
102 Shih, Pittinsky, & Ambady, 1999.
103 Strickler & Ward, 2004.
on this topic is extremely hard to obtain
because it would not be ethical to set up a
testing condition (the stereotype threat sit¬
uation) that was intended to lower the per¬
formance of one group of people relative
to another. Statistical methods for detect¬
ing stereotype threat in high-stakes exami¬
nations without doing this have been sug¬
gested, but to my knowledge they have not
been applied.104
All in all, it seems to me that the
chief social reason for discrepancies between
men’s and women’s performance in math¬
ematics, and the concomitant and perhaps
related discrepancy in visual-spatial reason¬
ing, is simply a difference in interests. This
difference causes girls to take fewer math¬
ematics courses than boys, on beyond the
primary grades. Because we learn what we
practice, adult women are less likely than
men to have acquired high levels of math¬
ematical skill. Due to differences in inter¬
est, women are less likely than men to pur¬
sue careers in the STEM fields, even when
they have developed the necessary cognitive
skills.
This conclusion leaves two questions
open: why do women develop different
interests than men, and are there biological
reasons for the discrepancy?
BIOLOGICAL REASONS FOR MALE-
FEMALE DIFFERENCES IN COGNITION
Evolutionary psychologists have offered a
widely held distal biological explanation for
differences in cognition between men and
women - the “man the hunter, woman
the gatherer” explanation. In every society
there are sex role differences that go beyond
the childbearing difference forced by biol¬
ogy. The evolutionary psychologists con¬
clude that these differences can be traced
to an evolutionary advantage held by groups
that, in prehistoric times, assigned the hunt¬
ing role to males and the gatherer-child care
role to females.
The story is that men in prehistoric soci¬
eties ranged widely as they hunted, and that
those men who were better hunters had a
104 Wicherts, Dolan, & Hessen, 2005.
THE DEMOGRAPHY OF INTELLIGENCE
reproductive advantage, either because their
success gave them more access to females,
or because their offspring were more likely
to survive to reproductive age because good
hunters could feed their children more reli¬
ably. This explanation also assumes that
skills related to the R dimension of the g-
VPR model, like the ability to judge the
velocity of moving objects or to maintain
orientation in space, aided in hunting. On
the other hand, women are supposed to have
primarily been gatherers of edible plants
and small animals near the camp. A woman
would have a reproductive advantage if she
were a good gatherer, and so better able
to feed her children. The final step is to
assume that the ability to notice fine details,
like an edible lizard in a bush or an edi¬
ble berry amid shrubbery, would make for
a better gatherer. Women’s superior ver¬
bal abilities are explained by the assump¬
tion that women, being more dependent
on others for protection of themselves and
their offspring, had to be superior in social
interchanges.105
The story is plausible. Slight reproductive
advantages associated with various skills,
acting over thousands of generations, could
produce a substantial sexual imbalance in
skills. However, the story is a story, not
a fact. There is no direct evidence for it,
because we know little, and probably never
will know very much, about the behavioral
characteristics of prehistoric Homo sapiens,
let alone other hominid predecessors of our
species. What we do know is inferred from
indirect evidence, such as inferences about
the rate of maturation of extinct hominids
based on skeletal data, which then are used
to infer the years required for protection
of children, and then extrapolated to dis¬
cussions of male-female roles in mainte¬
nance of children, foraging, and group pro¬
tection. Analogies to the behaviors of exist¬
ing human hunter-gatherer societies are fre¬
quently used.
Evolution certainly happened, and it
could have happened this way. But the evi¬
dence is certainly, and of necessity, indirect.
105 Geary, 2005, 2oo7a,b.
4° 3
There is another evolutionary story,
based on social rather than biological mech¬
anisms. The human brain is designed for
general learning, a trait that is very useful
in a species that occupies multiple ecologi¬
cal niches. Certain behavioral practices that
produce better social and economic organi¬
zation can lead to a competitive advantage
at the level of the group, rather than the
individual. Groups that adopt such prac¬
tices are more likely to survive.106 These
practices include differentiation of male and
female roles. Since humans are learners,
males and females will acquire different,
role-appropriate cognitive skills. The ten¬
dency will be accentuated over time, but the
accentuation will be due to social rather than
biological evolution. The difference in cog¬
nitive skills that appears in late adolescence
is due to different learning experiences that
have evolved historically, not physiologi¬
cal differences in brain structure that have
evolved genetically.
This hypothesis emphasizes social his¬
tory. It can be given a biological twist. It may
be that boys learn more spatial skills because
they explore more, and girls learn better
verbal skills because they socialize more.
This could be due to genetic influences on
the tendency to explore each environment.
Thus sex differences in cognition (and other
differences between people) could be due
to differential learning experiences, but the
differential learning experiences themselves
could be under genetic influence.107
I see no way to differentiate among these
hypotheses. In fact, they are not mutually
exclusive; all of them could have some truth.
It is tautologically true that all human vari¬
ations in behavior will be within the range
of variation permitted by the genome; the
most dedicated geneticist does not deny
that humans are superb learners. We have
learned to live with this ambiguity with
respect to physical behaviors. No one denies
that males are genetically predisposed to
be able to swing sticks more rapidly than
females, and no one denies that today’s
106 Campbell, 1975; Wilson & Wilson, 2007, 2008.
107 Bouchard, 1999.
HUMAN INTELLIGENCE
4°4
training methods routinely produce women
tennis players who hit the ball harder than
the men’s champions of yesteryear.
The strict version of the genetic hypoth¬
esis, that men and women have evolved to
have different cognitive abilities due to dif¬
ferent physiological structures, rather than
to different learning experiences, implies
that there may be some physical difference
between modern men and women that can
be related to cognitive differences. Has such
a physical difference been found?
Richard Lynn has pointed to one candi¬
date difference.108 Men have bigger brains
than women. After allowing for body size,
men’s brain volumes average about 100
cm3 larger than women’s brain volumes.109
Within the sexes, brain volumes are posi¬
tively correlated with intelligence test scores
(r ~ .35; see Chapter 7). Lynn argues that
the difference in brain size accounts for the
difference in general intelligence. He has
amplified this argument by observing that
girls mature faster than boys, so that the dif¬
ference in brain size does not appear until
after adolescence. This parallels the devel¬
opment of the claimed difference in general
intelligence.
The explanation is appealingly simple and
probably incorrect.
The evidence for any substantial differ¬
ence between men and women in gen¬
eral intelligence is extremely weak. There¬
fore, Lynn’s argument leads to a paradox
that, although not completely damning to
Lynn’s line of reasoning, certainly makes it
seem implausible. There are brain size dif¬
ferences of about 20 cm3 between African-
and European-derived populations in the
United States, with the Europeans hav¬
ing larger brains after correction for body
size. This is much smaller than the dif¬
ference in brain volume between men
and women. However, the difference in
test scores between African-derived and
European-derived Americans, roughly 15
points on the IQ scale, is four to five
times larger than any reported difference
108 Lynn, 1994, 1999; Colom & Lynn, 2004.
logAnkney, 1992
in test scores between men and women.
This poses a contradiction, for it is not clear
how a small brain size difference could pro¬
duce a large IQ difference between races,
while a large brain size difference produced
a small IQ difference between men and
women.110
The fact that the male-female brain size
discrepancy and the discrepancy in general
intelligence (if it exists) develop in paral¬
lel is not a strong argument for causation.
There are many changes in the difference
between girls and boys from childhood to
adolescence. These include both hormonal
changes that alter brain action and changes
in social roles and experiences. Picking out
one of many bivariate correlations and offer¬
ing it as the explanation for a phenomenon,
without a serious consideration of alterna¬
tive hypotheses, is not a compelling scien¬
tific argument.
MALE-FEMALE DIFFERENCES IN BRAIN
ORGANIZATION
A fairly strong case can be made for the exis¬
tence of male-female differences in brain
organization that are related to differences
in cognition.
Neural pathways can be divided into the
myelinated white matter pathways and the
unmyelinated gray matter pathways. The
white matter pathways connect relatively
distant centers of the brain, while the gray
matter pathways deal with local connec¬
tions. There is a loose analogy to the dis¬
tinction between highways and city streets.
Men have a higher proportion of white
matter to gray matter than women do.
In addition, women have somewhat larger
connections between the hemispheres. This
suggests that the male brain relies more
on specialized processing centers in one or
the other hemisphere than does the female
brain, while the female brain spreads pro¬
cessing out across several possibly less pow¬
erful centers that perform the same or
related functions. This is consistent with
the observation that women have better
110 Macintosh, 1998, p. 184.
THE DEMOGRAPHY OF INTELLIGENCE
prospects of recovering from localized brain
damage than men.111
These ideas have been supported by
imaging studies, which have shown that
when men and women are faced with the
same problems, such as solving progressive
matrix items, they show somewhat differ¬
ent patterns of activation, and that perfor¬
mance correlates with anatomical measure¬
ments, such as white/gray matter ratios, in
different places in the brain.112
The finding that men’s brains and
women’s brains are, on the average, active
in different places during problem solving
excited a good deal of comment in the press,
because the idea that “brain studies show
that men and women think in different
ways" understandably arouses a good deal
of popular interest. The finding should not
be overinterpreted. Brain activities during
problem solving will depend upon the
strategy that the problem solver takes. It
has been shown experimentally, and is not
logically too surprising, that an individual
can alter his or her brain activity by altering
his or her problem-solving strategy.113 It
does not follow that men and women must
adopt different problem-solving strategies
because their brains are different, although
it may be the case that differences in their
brains predispose men and women to adopt
different strategies. This conclusion does
not lead to sweeping generalities, but that
is the way it is!
Spatial orientation and navigation
deserve special attention. These aspects of
thought, part of the R in the g-VPR model,
produce some of the largest, most consistent
male-female differences in behavior. They
also produce an interesting qualitative
distinction. Men are more likely to orient
themselves in space using geometric cues
about the overall environment, while
women are more likely to use landmark and
route information.114 The same difference in
in See Gur & Gur, 2007, and Kimura, 1999, Chap¬
ters 10 and 11, for further references and discussion.
112 Haier, Jung, et al., 2004, 2005.
113 Reichle, Carpenter & Just, 2000.
114 Hunt, 2002, Chapter 6; Maguire, Burgess, &
O'Keefe, 1999.
4°5
behavior is found when comparing differ¬
ences in the way in which male and female
rats explore unfamiliar environments.11* The
rat behavior has been associated with neural
activation in the hippocampus and related
structures deep in the medial temporal
region of the brain. There are suggestions
of differences in hippocampal structures
between men and women, but the data is
not yet definitive. As in the case of problem
solving, orientation and navigation can
be achieved by different strategies, and
the strategy used will determine the brain
region involved.
To summarize, there are differences both
in the structures of male and female brains
and in the ways in which the brain is used
in different aspects of problem solving. The
differences seem to be especially large when
the problem to be solved has a spatial-visual
component. But what causes these struc¬
tural differences and differences in activity
in the first place? The answer to this turns
out to be genetic potential (of course!] inter¬
acting with hormonal balance during key
periods of brain development.
The evidence for this comes from two
sources: studies on nonhuman animals and
studies of abnormal human development.
Manipulation of adrenal levels prenatally
and postnatally can influence the display of
typical male or typical female behavior in
rats, including the extent of engagement in
rough-and-tumble play and the patterns of
behavior in maze exploration. Obviously,
conducting an analogous experimental study
on humans would not be ethical. What we
can do is to study certain medical conditions
in which unusual hormonal concentrations
occur.
Congenital adrenal hyperplasia (CAH] is
a genetic condition in which the adrenal
gland fails to generate a key enzyme, causing
unusual sensitivity to male hormones. The
condition can occur in both boys and girls. It
is treated by restoring the normal hormone
balance. Female CAH patients tend to have
higher scores on spatial orientation tests
than normal females. Males with CAH (a
115 Ohnishi et al., 2006.
406
HUMAN INTELLIGENCE
less-studied group) tend to have lower scores
than normal males. There are indications
that this result generalizes to other behavior
patterns in women, for female CAH patients
display more masculine behaviors and inter¬
ests than do normal girls and women, includ¬
ing such things as preferences for “male-
appropriate” or “female-appropriate” toys.116
These results refer to effects of hormones
on the developing brain. There is also evi¬
dence that circulating hormonal levels in
adults will influence human cognition. The
results are somewhat inconsistent, as the
studies are difficult to do and generally
involve small numbers of participants, a con¬
dition that invites null findings. Neverthe¬
less, certain results seem to be reasonably
well established.
In women, high levels of circulating estro¬
gens facilitate tasks involving verbal fluency
and/or short-term memory. The evidence is
mixed for performance on visual-spatial rea¬
soning, except for a consistent reduction in
performance on mental rotation tasks. This
has been established by two sources of data:
studies of women tested at various times
during their menstrual cycle, and studies
of post-menopausal women who either are
or are not receiving estrogen replacement
therapy.117
Complementary results have been found
in studies of testosterone. Testosterone ap¬
pears to have a non-monotonic effect on spa¬
tial reasoning, enhancing spatial reasoning
in women and men with low testosterone
(a common condition in the elderly), but
decreasing spatial reasoning abilities in men
with normal or high testosterone levels. The
cognitive effects are complicated by the fact
that circulating testosterone levels are asso¬
ciated with a myriad of other effects, includ¬
ing increases in impulsivity and aggressive
behavior. To get some idea of the com¬
plexity of the effect, contrasting effects of
testosterone on spatial rotation have been
reported in England, the United States,
116 Berenbaum & Resnick, 2007; Kimura, 1999, Chap¬
ter 9; Puts et al., 2008.
117 Haussmann et al., 2000, and references therein;
Halpern 81 Tan, 2001.
and China.118 The authors speculate that this
is because of different emphases on speed
versus accuracy of response in different cul¬
tures. The more general point is that in
spatial-visual problem solving (which can be
quite difficult) a hormonal effect could be on
either the brain mechanisms required for the
task itself or the brain mechanisms involved
in selecting a problem-solving strategy.119
11.3.7. A Summary but No Answers
There is no simple answer to the ques¬
tion “What causes male-female differences
in cognition?” There are biological differ¬
ences between men's and women's brains,
and these differences influence both verbal
fluency and, to a somewhat greater degree,
spatial orientation, way finding, and the
manipulation of visual images - the V and
R dimensions of the g-VPR model. How¬
ever, biology is not destiny. Just as there
is no doubt that there are biological pre¬
dispositions that operate to produce a femi¬
nine or masculine trend in thought, there are
social influences that can either accentuate
or override these trends. Given motivation,
humans are powerful learners, and learning
itself influences brain organization.
Confused? You should be. Just as Diane
Halpern said, the more you learn about
sex differences in cognition, the less certain
you are that you know the answers. Within
very broad limits, biological and social influ¬
ences produce tendencies toward certain
cognitive and social behaviors. Once this is
done, the behaviors themselves alter biolog¬
ical makeup and social status. Women, on
the average, are somewhat more verbal and
somewhat less spatially oriented than men,
but there are male chatterboxes and female
helicopter pilots.
There is one question that we can
deal with. Are men more intelligent than
women? The answer is clear.
It is the wrong question. Men and women
have somewhat different brains. This leads
118 Yang et al., 2007.
119 Kimura, 1999, Chapter 9.
THE DEMOGRAPHY OF INTELLIGENCE
to somewhat different strengths and weak¬
nesses in cognition. Any battery-type intelli¬
gence test will, by definition, consist of a mix
of tasks, some that favor the feminine style
of thinking and some that favor the mas¬
culine style. The particular mixtures found
in battery-type tests, such as the WAIS,
have been determined more by history than
by any analysis of why one mix is better
than another. There have also been power¬
ful commercial forces that work against try¬
ing out any test battery that is too different
from its predecessors. We should not disre¬
gard present batteries, and any male-female
differences built into them, but we should
not regard them as set in stone.
One could construct test batteries that
favor males, favor women, or are neutral.
There would be no way to say which of
these mixes was best without specifying the
purpose of the test. A test battery for the
selection of pilots will not and should not
look like a test battery for the selection of
law school students.
Any individual test, including the much-
used matrix tests, will evaluate g and test-
specific features, a point that Spearman
made a century ago. You can find a mea¬
sure of general intelligence that is g and ver¬
bally loaded, and produce an advantage for
females, or produce a measure of general
intelligence that is g and loaded on spatial-
visual reasoning, and find an advantage for
males. Depending on the purpose for which
the test is to be used, one or the other solu¬
tion could be the correct one.
It is sensible to try to understand male-
female differences in cognition. It is not sen¬
sible to ask whether men are smarter than
women.
11.4. Race and Ethnicity
This section will deal with racial and eth¬
nic differences in intelligence. The topic is
probably the most controversial one in psy¬
chology. Jerry Carlson and I identified the
following positions that have been taken:120
4°7
1. There are no differences in intelligence
between racial and ethnic groups.
2. There are large differences in intelli¬
gence between racial and ethnic groups.
3. Differences in intelligence between
groups exist and are social in origin.
4. Differences in intelligence between
groups are genetic in origin.
5. There is no such thing as race.
The facts concerning racial and ethnic dif¬
ferences in IQ and similar test scores are
clear. The causes and implications of these
facts are not at all clear. Therefore, the sci¬
entific discussion can easily slip over into a
heated dispute. The discussion and the dis¬
pute cannot be kept entirely apart. How¬
ever, I believe (hope) the discussion here
will shed more light than heat. To this end
I will first spend some time setting up the
problem and identifying the key variables,
prior to presenting a picture of what I think
is known. Being hyper-careful about the def¬
inition and scope of a subject may be some¬
what boring, but in this case precision of
thought is essential.
In previous discussions of the conceptual
nature of intelligence, beyond the opera¬
tional definition of a test score, the following
points were made:
1. Although the requirements for intelli¬
gence vary somewhat across societies,
there is a common core - for exam¬
ple, the ability to comprehend verbal
arguments or to maintain orientation in
space.
2. Intelligence tests and their analogs,
such as personnel classification tests,
are assessments of the cognitive skills
required for success in industrial and
post-industrial society. The test scores,
although not perfect indicators of intel¬
ligence, are relevant in that society.
They will be relevant to success in other
societies to the extent that the tests tap
the cognitive skills relevant for success
in the target society. The extent of rel¬
evance has to be determined on a case-
by-case basis.
120 Hunt & Carlson, 2007a.
408
HUMAN INTELLIGENCE
3. Evaluating the sorts of skills required
by the post-industrial societies is impor-
tant; in itself. These societies dominate
the globe now and will do so for the
foreseeable future. Learning about intel¬
ligence as it is expressed in London,
Madrid, or New York will have more
impact on the human condition than
learning about the expression of intel¬
ligence in small groups in the forests of
South America. This is not to say that
anthropological studies are uninterest¬
ing; they can be very informative. The
point is just that even if it were shown
that intelligence, as tested, is wholly
confined to the post-industrial societies,
the topic would still be important.
4. All cognitive skills are ultimately deter¬
mined by biology. However, it is impor¬
tant to distinguish between genetic con¬
tributions, which establish potentials,
and life experiences, which determine
the extent to which a potential is real¬
ized. This can be very difficult to do.
Measuring the size of the genetic contri¬
bution to variance in a particular society
or subsociety is not the same as under¬
standing it.
Do races and ethnic groups exist? It has
been claimed that these are solely social dis¬
tinctions. Because the definition of a group
varies from time to time and place to place,
it is argued, they cannot be biological nat¬
ural kinds. Therefore, racial differences are
not amenable to scientific study because the
basic concept of race is ill-defined.121 This
argument has been buttressed by the fact
that genetic diversity within groups is larger
than genetic diversity between groups.122
The counterargument is that in fact, peo¬
ple identify and are identified with racial and
ethnic groups. These distinctions are highly
correlated with a clustering of social and
biological variables. Socially, for instance,
68% of the residents of the United States
who identify themselves as Hispanic are
Catholics. Only 20% of the non-Hispanic
121 Fish, 2002; Smedley & Smedley, 2005.
122 Lewontin, 1974.
residents are Catholic.123 Therefore, Catholi¬
cism is statistically associated with ethnic
identity, even though there are Hispan-
ics who are not Catholic and Catholics
who are not Hispanic. The same reason¬
ing applies to genetic variation. While it is
true that within-group genetic variation is
greater than between-group variation, the
amount of genetic variation between groups
is quite sufficient, statistically, to make accu¬
rate assignments associating an individual
with one of the three largest groups of ori¬
gin in the United States - African, European,
and East Asian.124 There is also a high level
of agreement between self-identification as
White, Black, East Asian, or Latino and
assignment of a person to clusters based
upon similarities of their genomes.125 While
there is no one defining characteristic, other
than self-identification, that assigns some¬
one to a particular racial/ethnic group, iden¬
tification of a person as a member of a
racial/ethnic group will probabilistically tell
us something about that person’s standing
on a variety of social and biological variables.
The key word is “probabilistic.” Artifi¬
cial Intelligence researchers have developed
the idea of a “fuzzy” concept, a class in
which membership is determined to varying
degrees rather than being a binary property.
The political concept conservative is a good
example. An individual may espouse con¬
servative beliefs with an intensity that varies
with the topic. Race and ethnic member¬
ship are fuzzy concepts. There are no defin¬
ing characteristics of a racial or ethnic group,
but there are statistical tendencies.
The de facto definition of races and eth¬
nic groups - that is, the extent to which
a person has to possess the (fuzzy) prop¬
erties in order to be designated a group
member - varies over time and place. In
modern American categorization the term
“Asian- American” includes genetic Japanese
who are fourth-generation Americans, who
do not speak Japanese and have never been
to Japan, along with green card holders
123 New York Times, 26 April 2007.
124 Bamshad et al, 2004.
125 Tang et al., 2005.
THE DEMOGRAPHY OF INTELLIGENCE
409
who just landed from Cambodia and speak
marginal English. Similar examples could be
offered for every other large racial and eth¬
nic group in North America and Europe.
Such variety means that for some purposes
the classification of individuals into broad
racial/ethnic groups is nearly useless. For
other purposes it makes sense.
Knowledge about covariation between
racial/ethnic identification and important
social outcomes, such as health status, can
serve as clues to tell us what variables to
look at as we attempt to understand causal
relationships. For instance, it is empirically
the case that high blood pressure is more
common in the African American than in
the White community.126 What is it in the
African American genetic constitution or
social practice that leads to high blood pres¬
sure? Questions like this ought to be inves¬
tigated by the biomedical sciences.127 Simi¬
larly, if there is a discrepancy in intelligence
test scores between the African American
and other American communities (and there
is), seeking to understand the reasons for
and implications of the discrepancy are legit¬
imate issues for the psychological sciences.
Research on racial/ethnic distinctions can
inform policy making. Most of the devel¬
oped societies do make policy decisions
about racial/ethnic groups. The United
States, for instance, has developed an elab¬
orate legal doctrine aimed at the prohibi¬
tion of discrimination. Two American psy¬
chologists, A. & B. D. Smedley, writing
in the American Psychologist , a high-impact
journal, vehemently rejected the idea of
race as a biological concept, and at the
same time urged the government to con¬
tinue to compile statistics on social inequal¬
ity between races (and, by extension, ethnic
groups) for the purpose of eliminating such
inequalities.128
Such a position contains a contradiction.
Policy makers should be informed of rel¬
evant scientific information, which should
126 Information downloaded from www.americanheart
.org/presenter.jhtml?identifier=462i, January 2009.
127 See Bamshad et al., 2004, for further examples.
128 Smedley & Smedley, 2005.
be considered along with other things when
policy decisions are made. In the case of
racial/ethnic differences, the scientific infor¬
mation needed can be obtained only by doc¬
umenting the existence of ethnic differences
in intelligence, understanding the implica¬
tions of these differences, and inquiring into
all possible causes, both social and biologi¬
cal. The Smedleys worried, with some rea¬
son, that such information could be used to
justify discriminatory practices. This could
happen; but if it does, it will be the result of
political decisions combined with scientific
findings, not the scientific findings them¬
selves. I strongly believe that it is better
to proceed with information rather than to
proceed in ignorance.
In practice, investigators who study racial
and ethnic differences face major obstacles.
These obstacles have had an impact on the
quality of research in the field. Let us see
why.
11.4.1. Issues of Quality in Research
on Race and Ethnicity
Racial and ethnic distinctions are both
fuzzy, in the sense just defined, and incon¬
sistent over time and place. Brazilians intro¬
duce distinctions that Americans do not,
referring to people as white and black, but
also as Moreno, mestizo, and pardo. The
terms Indian and Native American are used
today much more broadly than they were
in the past. Given that the answer to the
question “who is what?” changes over place
and time, it is no wonder that it is hard to
find how a behavioral capacity, such as intel¬
ligence, correlates with race and ethnicity.
Race and ethnicity are seldom causal vari¬
ables in themselves. The only exceptions are
in studies of the reactions of others to a
person of a given race, and studies of the
effects of self-identity on people’s concep¬
tions of themselves. In these situations the
application of a label may itself be a stim¬
ulus that leads to action. Outside of such
situations, a researcher who studies a corre¬
lation between racial/ethnic status and a test
score is studying the correlation between the
score and the many possibly causal variables
410
HUMAN INTELLIGENCE
correlated with race and ethnicity. These
include educational status, biological dis¬
tinctions, health and wealth, and a variety of
beliefs and behavioral practices. This com¬
plicates the assignment of causes for indi¬
vidual differences. The bivariate argument
that “because racial/ethnic groups differ on
possible cause X and outcome Y, cause X
must produce group differences on Y” is an
example of seizing on a bivariate correla¬
tion in the presence of collinearity. Such
arguments should be regarded with great
suspicion.
Differential recruitment is also a prob¬
lem. In particular, college students, the
default sample of convenience, are easy to
study. However, college students are not a
random sample of any racial or ethnic group,
which makes generalizations tricky.
Finally, there are social (and sometimes
legal) pressures restricting research on racial
and ethnic groups. While I may think that
the need to overcome ignorance overrides
the possibility that a finding on race or eth¬
nicity may be used to justify discrimination,
there are other opinions. There are audi¬
ences who believe that anyone who reports
racial/ethnic differences must be a racist,
and concomitantly that anyone who reports
a finding questioning the existence of racial
and ethnic differences in intelligence must
be on the side of the angels. There are also
audiences who are longing to hear about
racial differences in intelligence, because
they want to use such findings to justify dis¬
criminating against, or not providing affir¬
mative action for, an ethnic group. This has
led both to a difficulty in getting funds to do
good research in the field and to a tempta¬
tion to oversimplify a finding, in order to tell
different audiences what they want to hear.
I am not alone in my concerns. In the
1970s Lindzey, Loehlin, and Spuhler closed a
widely respected review of racial difference
in intelligence by saying
The design , execution , and reporting of
studies of racial- ethnic differences in intel¬
ligence often leave much to be desired.
The conclusions that we have attempted
to draw from these data are often limited
by this fact. We have been concerned
privately by the number of instances in
which the political and social preferences
of the investigators apparently have grossly
biased their interpretations of the data.
Such distortions appear to be at least as
prevalent at environmentalist as at heredi-
tarian extremes.
Lindzey, Loehlin, & Spuhler, 1975,
pp. 232-233
Writing over thirty years later, I am sorry
that I have to say that Loehlin, Lindzey,
and Spuhler are still correct. Bad research
is itself a fuzzy concept. Very few studies
are uniformly good or bad. In this field, more
than in others, it is important to consider the
strengths and weaknesses of a study before
generalizing its findings. Panel 11.1 presented
some cautions about research on group dif¬
ferences. I urge readers to consider their
impact when evaluating a study on racial and
ethnic groups, reported here or anywhere
else.
11.4.2. Coverage and Terminology
For the most part, I will be concerned with
the four major racial/ethnic groups in the
United States, African Americans, Asian-
Americans, Hispanics, and Whites. On occa¬
sion studies from other countries will be
considered. These are the groups for which
the most data has been gathered. In addition,
these groups live in a large, post-industrial
society; the vast majority of their members
speak a common language, English, either as
a first language or a strongly held second lan¬
guage; and they have generally comparable,
although certainly not identical, living stan¬
dards, compared to the contrast between,
say, Whites in the United Kingdom and
Africans in Somalia.
The four American groups have all the
characteristics of a fuzzy class, defined on
either social or biological variables. There is
no one variable, or even one group of vari¬
ables, that defines group membership. How¬
ever, statistical clusterings based on either
social or genetic variables do correspond
closely to self-identification with a group.
In addition, these groups have a political
identity. Laws, regulations, and government
THE DEMOGRAPHY OF INTELLIGENCE
411
programs exist that distinguish among them
in a variety of ways. As a consequence,
the government keeps extensive statistics
on characteristics of their health, education,
and economic status. It is often important to
consult these statistics in order to untangle
issues due to collinearity.
Finally, a word about terminology.
Anyone who attempts to name racial/
ethnic groups in the United States is shoot¬
ing at a moving target. W. E. B. Du Bois
(1868-1963), a leading social activist of the
1900-50 period and one of the founders of
the National Association for the Advance¬
ment of Colored People (NAACP), referred
to himself as a Negro. The term is proscribed
today. The term “colored people” is part of
the NAACP's name, but is now also pro¬
scribed. However, the term “people of color”
is appropriate, at least as of 2010.
I will use the terms African American and
Black interchangeably unless, of course, I
am referring to a group outside the United
States. In present US government docu¬
ments Hispanic or Latino is used to refer to
US residents who are immigrants from, or
descendants of immigrants from, a Spanish-
or Portuguese-speaking country in the West¬
ern Hemisphere. The term is also used for
descendants of the Spanish/Mexican people
who settled in California and the South¬
west prior to the 1840s. Surprisingly, the
government does not include in this term
people whose family origin was Spain or
Portugal, although this policy seems to be
applied inconsistently. I will follow govern¬
ment usage, as it exists in 2010.
While I shall have to use the term Asian-
American at times, especially when refer¬
ring to government records, I will attempt
to refer to more specific groups, such as the
Japanese, as I believe that the current official
designation Asian- American is far too broad.
White will be my catch-all term for all
other groups. These are primarily Ameri¬
can residents of European descent. There are
substantial communities who have cultural
and genetic ties to Armenia, Georgia, Iran,
Israel, and other Middle Eastern nations, but
they are all European- Americans to the US
census! When appropriate I will refer to
specific communities within the broad Asian
and White designations.
11.4.3. Test Score Gaps between Whites ,
African Americans , and Hispanics
There is a long history of studies of racial/
ethnic differences in test scores. In their 1975
review Lindzey, Loehlin, and Spuhler, citing
earlier data by Yerkes, reported that in 1917-
18, during World War I, the mean score on
the Army tests for White recruits was 1.16
standard deviation units above the mean for
African American recruits. Studies of enlis¬
tees in World War II and the Vietnam War129
showed a 1.52 mean difference in favor of
Whites. Lindzey and colleagues were care¬
ful to point out that this is not evidence that
the Black- White differences in intelligence
had increased from 1918 to the 1960s, because
military enlistees are not a representative
sample of the country, and because different
recruitment/conscription standards were in
effect in the two wars. However this is cer¬
tainly not evidence for a presumed reduction
in the difference!
In order to make a comparison between
the scores of different groups we need to
have data from a representative sample of
the national population. Table 11.4 presents
the results from several such surveys involv¬
ing battery-type tests. There is some vari¬
ety in the results, but not a great deal. The
African American means are about 1 stan¬
dard deviation unit (15 points on the IQ
scale) below the White means, and the His¬
panic means fall in between.
A similar picture is obtained from
comparisons involving the Raven Pro¬
gressive Matrices (RPM) tests. Figure
11.21 shows the median RPM test score
obtained in a school district in the west¬
ern United States, as a function of age
and racial/ethnic group.130 We see the
same picture reflected in the scores on
battery-type tests. Whites outscore African
129 Lindzey, Loehlin, & Spuhler, 1975, p. 143.
130 Raven, 2008a. J. Raven has recommended reporting
RPM scores as percentiles, rather than in terms of
summary statistics, such as means and variances. See
Raven, 2008b, p. 60 (note 1.55), for a justification.
412
HUMAN INTELLIGENCE
Table 11.4. Values of (White mean - African American mean) and (White mean - Latino
mean) in standard deviation units for a variety of cognitive tests, using cases where a
reasonably large standardization sample has been obtained
African
Population
Test Used
American
Latino
Reference
WAIS III adult
standardization
WAIS General
Ability Index
•95
■65
Lange et al., 2006
NLSY79, men
AFQT
1.07
Scullin et al., 2000
NLSY79, women
AFQT
1.00
Scullin et al., 2000
NLSY79, all groups
AFQT
1.2
•93
Herrnstein &
Murray, 1994,
pp. 275, 278
Standardization sample
Wide Range
Intelligence Test
.85
■51
Shields, Konold,
8c Glutting, 2004
Standardization sample
Woodcock-Johnson 3:
General Intellectual
Ability
1.05
Murray, 2007
Americans, and Hispanic scores fall some¬
where in between.
In order to avoid recruitment effects,
Table 11.4 cites studies using relatively
large samples, where an attempt was made
to obtain a sample representative of a
defined population. The samples involved
cover a wide range of people, from the
five to sixty-five years age range in the
Woodcock-Johnson standardization sample
to the schoolchildren studied in the RPM
standardization. Similar results can be found
by averaging over the more than 150 studies
that have used convenience samples.131
Similar differences are found internation¬
ally. Historically there have been a num¬
ber of studies comparing Whites to other
racial/ethnic groups in a variety of countries.
Because there have been major changes in
the economic and health status of many
developing countries, the best course is
probably to look at the recent literature
rather than at that of over thirty years ago.
J. P. Rushton, Lynn, and a number of
their colleagues have conducted a wide-
131 Herrnstein & Murray, 1994, p. 277; Jensen, 1998,
p. 354. The Herrnstein and Murray citation gives
references to specific studies. Jensen’s citation does
not, but it apparently refers to an analysis that he
conducted.
ranging series of studies in which they
use the Raven Progressive Matrices tests
to evaluate group differences within var¬
ious countries. All obtain the general
results observed on the Raven standardiza¬
tion. Whites do better than Blacks, with
other ethnic groups somewhere in between.
The studies involved include a Roma-
Serbian contrast in the Balkans/32 White-
Indian-mixed race-Black contrasts in South
Africa/33 and a contrast between Whites,
Mestizos (mixed White-Native American),
and Native Americans in Mexico.134 In all
these studies Whites and Asians obtain
the highest scores, and Blacks the lowest,
with other racial/ethnic groups falling in
between.
The Mexican study provides a good
example of this work, because it is some¬
what more extensive than several of the
other studies. Lynn and colleagues tested
elementary school children, aged seven to
ten, near Ensenada, in the state of Baja Cali¬
fornia. The ordering of means was what the
experimenters had anticipated: Whites (IQ
equivalent ~ 100), Mestizos (mixed Native
132 Rushton, Cvorovic, & Bons, 2007.
133 Rushton & Skuy, 2000; Rushton, Skuy, & Fridjohn,
2003.
134 Lynn, Backhoff, & Contreras, 2005.
THE DEMOGRAPHY OF INTELLIGENCE
4*3
Figure 11.21. Raven Progressive Matrices scores in a school district
in the western United States, as a function of age and racial/ethnic
group. Data taken in the 1980s. Source: J. Raven, 2008b, Table 8.3.
American-White groups) (IQ equivalent ^
95), and Native Americans (IQ equivalent^
83). (See Figure 11,22.) The authors point out
that these results resemble those obtained
in the United States, where Mexican immi¬
grants have scores below Whites, and Native
Americans tend to have still lower scores.
The differences were substantial. Seven-
year-old White and Mestizo children solved
progressive matrix problems at a level not
obtained by Native American children until
they are nine or ten.
Lynn and colleagues’ results for Mexi¬
cans in Mexico can be compared to Raven’s
results for Latinos in the southwestern
United States. This is done in Figure 11.23,
which shows a striking continuity in changes
of test scores across ages, within each ethnic
group.
Because the Raven tests are often referred
to as measures of g, there is a temptation
to interpret these results as showing that
White populations possess general intelli¬
gence to a greater degree than nonwhite
populations living in close proximity to the
White groups. As Lynn himself has indi¬
cated, it is not appropriate to draw such
a conclusion based on results from a sin¬
gle test. However, similar results implicat¬
ing differences in g have also been found
in European studies that used batteries of
subtests designed to evaluate narrow cogni¬
tive functions. The populations compared
included children from different immigrant
Figure 11.22. The median number of Raven's Standard Progressive
Matrices problems solved by Native American, Mestizo, and White
children in Mexico. Data from Lynn, Backhoff, & Contreras, 2005,
Table 2.
44
HUMAN INTELLIGENCE
6 7 8 9 10 11 12 13 14 15 16 17
Age of Children
— *— Lynn White
Raven White
Lynn Mestizo
Raven Hispanic
Figure 11.23. Median number of problems solved as a function of
age and ethnicity. Data from Raven, 2008a, Table 8.2 (half-year
intervals) and from Lynn, Backhoff, & Contreras, 2005, Table 2.
groups in Europe and adult immigrants from
the Netherlands Antilles compared to native
Dutch employees in the railway system in
the Netherlands.135
11.4.4. ^ Closer Look at the Nature
of Racial/Ethnic Differences
In the middle of the twentieth century a
study was done on variations in first-grade
children’s intelligence that were associated
with ethnic status and social class (SES).136
The authors concluded that the level of
intelligence was associated with SES, and
that there were patterns of differential abil¬
ity associated with ethnic groups. Asians
were said to have high spatial ability, and
Jewish children to have high verbal abili¬
ties. Since that time there have been sev¬
eral efforts to determine the nature of the
differences in the intelligence of various
racial/ethnic groups, beyond the omnibus
statement that test scores vary from group
to group.
Most of these studies have been presented
as investigations of “Spearman’s hypothesis.”
The strong version of this hypothesis is that
all intergroup differences in intelligence are
due to differences in general intelligence.
The weak form is that the majority of these
differences are due to general intelligence,
135 Helms-Lorenz, van de Vijver, & Poortinga, 2003; te
Nijenhuis et al.( 2004.
136 Lesser, Fifer, and Clark, 1965.
but that differences in lower-order factors
(e.g., verbal and spatial- visual reasoning)
may also contribute to group differences.137
Jensen summarized a number of studies
testing Spearman’s hypothesis that had been
carried out through about 1995 using the
method of correlated vectors. He concluded
that the correlation between test loadings on
a g factor and the Black- White differences
in test scores is about .60, and considerably
higher in some tests.138 Jensen interpreted
this as substantial support for the weak form
of Spearman’s hypothesis. He further con¬
cluded that spatial-visual reasoning, which
tends to show fairly large Black- White dif¬
ferences, was responsible for the remaining
differences.
Subsequently the Danish psychologist
Helmut Nyborg collaborated with Jensen on
a large study of African American- White
differences among Vietnam veterans.139
Because conscription was used during the
Vietnam War the study participants were
somewhat representative of men from the
cohorts born in the United States dur¬
ing the 1940s and early 1950s. Nyborg and
his colleagues updated Jensen’s review and
published further data testing a White-
Hispanic contrast in the Vietnam veteran
population.140 These studies also reported
137 Jensen, 1998, p. 372.
138 Ibid., pp. 376 ff.
139 Nyborg & Jensen, 2001.
140 Hartmann, Kruuse, & Nyborg, 2007.
THE DEMOGRAPHY OF INTELLIGENCE
4*5
support for the weak form of the hypothesis,
that racial/ethnic differences in cognitive
skills are generally due to differences in g,
but that other traits may be involved.
Nyborg and Jensen’s conclusions, and
many similar conclusions about differences
between racial/ethnic groups, rest on the
validity of the method of correlated vectors.
Several questions have been raised about the
technique. These are described in Panel 11.6.
One of the most serious criticisms is that
the method of correlated vectors tests a sin¬
gle hypothesis against “chance,” the possibil¬
ity that the correlation between g loadings
and group differences would be produced
by random assignment of numbers. A bet¬
ter question is whether Spearman’s hypoth¬
esis fares well against competing hypothe¬
ses. Where a comparison has been made, the
results have been equivocal. In the study of
immigrant children cited earlier, the investi¬
gators also applied the method of correlated
vectors, but instead of using g loadings they
used “cultural loadings" assigned by having
a panel of graduate students rate whether
or not particular subtests were culturally
loaded. Aggregated cultural ratings had a
markedly higher correlation with group dif¬
ferences than the g loadings did.141
The conclusion that racial/ethnic group
differences are primarily due to differences
in g has been repeated several times in
the secondary literature, in spite of the
questions raised by Dolan’s analyses. The
Danish investigators who summarized the
evidence after Jensen’s 1998 summary did
not even mention Dolan’s work. Some¬
times the summarizers have actually mis¬
stated Dolan’s conclusions! When Rushton
and Jensen published what they regarded as
a summary of fifty years of studies of White-
Black differences, they had this to say about
Dolan’s work:
The results statistically confirmed the con¬
clusion derived from the method of cor¬
related vectors regarding a " weak form”
141 Helms-Lorenz, Van de Vijver, & Poortinga,
2003, Table 7. See particularly the lower right-
hand entries, which summarize the complicated
analyses.
of Spearman’s hypothesis: Black-White
group differences were predominantly on
the g factor , although the groups also
showed differences on some lower order fac¬
tors (e.g., short-term memory and spatial
ability) independent of g.
Rushton Jensen, 2005, p. 248
Here is what the original authors said to
summarize their work:
It is concluded that the Spearman correla¬
tion, as a test of the importance of gin B-W
differences, lacks specificity. The results of
the MGCFAs suggest that it is very difficult
to distinguish between competing hypothe¬
ses concerning the latent sources of B-W
differences.
Dolan Sc Hamaker , 2001, abstract
Where does this somewhat technical
argument leave the substantive conclusion?
Rushton and Jensen say that Spearman’s
hypothesis should be regarded as an empir¬
ical fact. The technical arguments and the
various demonstrations by Dolan and his
colleagues show that this is not the case.
However, a demonstration of the weakness
of the method should not be used to take
another (indefensible) position - that group
differences are independent of g. The results
of the Dutch study that contrasted cultural
and g differences as explanations for dif¬
ferences in test scores, although interest¬
ing, are in disagreement with several smaller
studies from the United States. However,
it is difficult to compare the degree of cul¬
tural difference between Native Dutch and
immigrants, on the one hand, and between
Whites and Blacks in the United States on
the other.
It is probably the case that most group
differences in cognitive skills are due in part
to differences in general reasoning, but the
extent is not clear. It would be nice to have
a reanalysis of some of the key studies using
appropriate statistical methods.
11.4.5. Group Differences on Educational
Assessments
In the United States African American
students have markedly lower educational
achievement than Whites; Asian-Americans
416
HUMAN INTELLIGENCE
Panel 11.6. The James Watson Affair
James Watson (1928-) is arguably Amer¬
ica’s most famous living scientist. In 1962
he, along with his colleague Francis Crick,
received the Nobel Prize for their dis¬
covery of the structure of the DNA
molecule. Their work sparked the explo¬
sion of discoveries in molecular biology
that have continued to this day. Subse¬
quently Watson left his appointment at
Harvard to head the Cold Spring Harbor
(New York) Laboratory, which he built
into a major scientific institution. Late in
his career Watson was a leading figure in
the successful effort to catalog the human
genome.
In 2007 Watson, who has always been
an outspoken person, wrote an autobi¬
ography, Avoid Boring People . He was
scheduled to give several public talks in
England. Just prior to the talks he gave
an interview to the (London) Sunday
Times.* The following quotes are taken
verbatim from the article:
1. He is inherently gloomy about the
prospect of Africa.
2. All our social policies are based on the
fact that their intelligence is the same
as ours, whereas all the testing says not
really.
3. His hope is that everyone is equal, but
" people who have had to deal with
black employees find that this is not
true. ”
4. There is no firm reason to anticipate
that the intellectual capacities of peo¬
ples geographically separated in their
evolution should prove to have evolved
identically.
5. Our wanting to reserve equal powers
of reason as some universal heritage
of humanity will not be enough to make
it so.
A huge public furor ensued. The New
York Times published a lengthy discussion
of subsequent debates between Watson’s
supporters and critics. For some reason
this appeared on the Arts page!' Many
of Watson's speaking engagements were
cancelled abruptly. He was almost imme¬
diately removed as director of the Cold
Spring Harbor Laboratory.
Let us look more closely at what Wat¬
son said. Statement (1) is a personal opin¬
ion that is shared by many, many knowl¬
edgeable observers. Statement (2) is
probably a statement of fact. I know of
no announced policy for various “Aid to
Africa" programs that assumes that the
intelligence of Africans is inherently and
permanently limited. At the same time,
many aid organizations have a realistic
view of the state of the resources and
teacher training in African schools. Intel¬
ligence test scores are generally lower
in Africa, as is documented in the text
of this chapter. Statement (3) is also
a statement of fact. As references in
this chapter have shown, in the United
States the work performance evaluations
of African Americans are, on the aver¬
age, lower than the evaluations received
by white workers. This is true for both
objective and subjective evaluations. The
difference is much less than the differ¬
ence in test score averages. However,
Watson's statement was certainly worded
in a provocative manner. Statement (4)
follows from well-established models of
evolution and genetics. Populations sepa¬
rated in space, for many generations, and
subject to somewhat different selection
pressures will develop genetic distinc¬
tions, within the limits of genetic variabil¬
ity that permit viable phenotypes. This is
as true for intelligence as for any other
genetically influenced trait. Statement (3)
is also a fact. Several Walt Disney movies
notwithstanding, wishing for something
does not make it so.
In 2008 I was privileged to attend
a three-day seminar on “improving the
brain" that had been arranged by the Cold
THE DEMOGRAPHY OF INTELLIGENCE
4*7
Spring Harbor Laboratory in honor of
Watson’s retirement. Watson discussed
the incidents recounted here, with obvi¬
ous bitterness. He felt that his remarks
had been taken out of context.
The incident illustrates the extreme
sensitivity of any discussion of racial, eth¬
nic, or national differences in intelligence,
especially if it is linked to a discussion
of genetics. Watson's remarks, as printed
(and the London Times has stood strongly
behind their accuracy), seem to have
been rather like the famous Rorsach ink
blot test in clinical psychology. The peo¬
ple who dropped him from their speaking
engagements read a racist interpretation
into the remarks. I would want to have
heard the entire interview before 1 made
such an interpretation.
Given such strong sentiments, it is
hardly surprising that policy makers,
including many funders of scientific
research, have backed away from inquir¬
ing about racial/ethnic differences in
intelligence. This has inhibited inquiry
and rational discussion, especially by
those who hold an intermediate position
on heredity - environment issues.
* October 14, 2007.
1 New York Times, December 1, 2007, p. A17 and
following.
have equivalent or slightly higher achieve¬
ment; and, depending somewhat on the par¬
ticular Hispanic group involved, Hispanic
students’ achievement falls between those
of Whites and African Americans. The best
data relevant to this question comes from
the NAEP testing program, for an attempt
is made to make the NAEP samples repre¬
sentative of the population of the US K-12
system, excluding those students who have
significant mental or physical problems, and
those students who are not proficient in
English.142 The effect is often referred to
as the ' gap” in white-minority educational
achievement. Figure 11.24 shows the size of
the gap for seventeen -year-olds during the
1990-2010 period, and, for comparison, the
gap in the 1970s. The gap has been constant
since roughly the 1990s, but is substantially
smaller than it was in the 1970s. The narrow¬
ing of the gap is due to an increase in the per¬
formance of African American and Hispanic
students. This probably reflects the benefits
of the considerable national effort that has
142 Grissmer, Flanagan, & Williamson, 1998. The exclu¬
sion is appropriate because the purpose of the
NAEP is to track educational progress in typical
American schools. NAEP results will understate
the gap in the US population to the extent that
African American and Hispanic students are over¬
represented in the excluded groups.
been invested in improving minority educa¬
tion since the 1960s.
The drop in the educational gap over
time has been mirrored by a similar drop
in the gap between African Americans and
Whites on cognitive tests less tied to the edu¬
cational curriculum. Charles Murray ana¬
lyzed scores from the various standardiza¬
tions of the Woodcock- Johnson (W-J) test
battery that took place in 1976-77, 1986-
88, and 1996-1999. Recall, from Chapter 2,
that this test is designed to provide sep¬
arate estimates of Gc and Gf. Unlike the
educational data, the standardization sam¬
ple for the W-J tests included people over
a wide range of ages, and therefore birth
cohorts. Murray concluded that the gap
in W-J scores dropped substantially in the
cohorts born from i960 to 1970, and then sta¬
bilized at about .85 deviation units in subse¬
quent years. This trend mirrors the change
in educational achievement scores, for the
cohorts born from the late 1950s through
the early 1970s would be the seventeen-year-
olds tested during the period that the NAEP
gap was being reduced, and the cohorts born
from 1970 onward reached the age of seven¬
teen during the period that the NAEP gap
was stable.
The gap in educational achievement has
considerable social importance, because of
418
HUMAN INTELLIGENCE
320
2005 2002
1998 1994
Year of Testing
1992 1975
320
a
5 310
</)
2 300
| 290
<u
| 280
2
c 270
ra
a>
5 260
a.
< 250
Z
240
2008 2004 1999 1996 1994 1992 1978
Year of Testing
■ White
□ Black
□ Hispanic
■ White
□ Black
□ Hispanic
Figure 11.24. Mean NAEP scale scores for White, African American
and Hispanic seventeen-year-olds. Material downloaded from
nces.ed.gov/nationsreportcard/lttdata, October 2009.
the emphasis on education in the post¬
industrial workplace. Therefore, it is worth
examining the gap itself and its changes in
some detail.
After conducting extensive analyses of
a variety of national educational assess¬
ments, Hedges and Nowell concluded that
the reduction in the gap was largely due
to a reduction in the proportion of African
American students in the lowest ranks of the
test scores. They did not find an increase in
the proportion of African Americans in the
highest ranks.143 Consistent with this posi¬
tion, the achievement gap in the NAEP is
mirrored by a similar gap in SAT scores,
which indicates group differences in cog¬
nitive skill in the self-selected, but socially
important, subset of students who intend to
obtain education beyond high school (Figure
11.25). This is worrisome for those concerned
143 Hedges & Nowell, 1998, pp. 159-161, and their
Tables 5-3 and 5B-2.
about group equality of rewards, for social
and economic rewards are associated with
higher levels of cognitive skills (see Chap¬
ter 10).
Although the White-Black test score gap
is present in test scores of very young chil¬
dren, it widens markedly over the school
years. This can be used to argue that either
(a) African American children are not well
prepared for school or (b) the schools are
selectively failing to educate African Amer¬
ican children. A good deal of light was
shed on this issue in a tightly argued paper
by Meredith Phillips, of the University of
Chicago, and two of his colleagues, James
Crouch and John Ralph.144
On the average, Black students enter ele¬
mentary school with fewer of the skills
required to benefit from formal instruction
than do White students. Phillips and col¬
leagues pointed out that it is important to
144 Phillips, Crouse, & Ralph, 1998.
THE DEMOGRAPHY OF INTELLIGENCE
4*9
White African American Hispanic
Figure 11.25. White, African American, and Hispanic mean scores
on the SAT. The within-group standard deviation for this test is
slightly over 100, making the White-African American gap
approximately one d unit.
distinguish between the possibility that
Black students fall further and further
behind as they progress through the schools
from the possibility that all students, of
whatever race or ethnicity, who are ini¬
tially poorly prepared fall further and fur¬
ther behind.
Figure 11.26 presents the results of some
of Philips and colleagues' analyses of the dif¬
ference between Black and White achieve¬
ment in reading and mathematics. The fig¬
ure shows the size of the gap as computed
directly from test scores at the beginning
of second grade and at the end of high
school, either with or without corrections
for prior measures of achievement; on enter¬
ing school (second graders, corrected for first
grade scores]; and measures taken in the
ninth grade and earlier (twelfth graders, cor¬
rected for scores through the ninth grade).
The size of the achievement gap was sub¬
stantially reduced, although not quite elimi¬
nated, by considering the effects of the prior
scores. Evidently differential reactions to
schooling associated with race per se are only
a small part of the causal factors that lead to
a Black- White gap in student achievement.
Of course, this does not reduce the seri¬
ousness of the situation, but it does cast it
in a different light. The effect of schooling
Math., no controls 12
Math., no controls 2
Math, i prior scores 12
Math. | prior scores 2
Reading, no controls 12
Reading, no controls 2
Reading | prior scores 12
Reading | prior scores 2
White mean - Black mean, d units
Figure 11.26. Difference in deviation score units between Black and
White students' scores on achievement tests taken in the second
and twelfth grades. The effect is shown with and without controls
for differences between students in reading and mathematical skills
assessed earlier. Data from Phillips, Crouse, & Ralph, 1998, Tables
7-4 and 7-5-
420
HUMAN INTELLIGENCE
is certainly to increase cognitive skills. In
the process of increasing those skills school¬
ing increases the range of variation between
those who were or were not initially well
prepared for schooling.
11.4.6. Group Differences in Cognitive
Skills in the Workplace
Three sources of data are available to
study group differences in workplace per¬
formance: the test scores obtained by people
applying for jobs, supervisors’ ratings of on-
the-job performance, and objective indices
of performance. One can argue that the best
measures of workplace performance would
be the objective indices, that the supervi¬
sor’s ratings would be next best, and that the
test scores would be only indirect evidence
of performance. Unfortunately the availabil¬
ity of the data produces the opposite rank¬
ing; it is fairly easy to study test scores and
relatively expensive to obtain objective per¬
formance measures.
A great many of the studies of applicant
test performance use data from the Wonder-
lie Personnel Test [WPT) and its current ver¬
sion, the WPT-R. As a brief reminder, peo¬
ple who take the WPT have to switch very
rapidly from doing one simple task [sim¬
ple mathematics word problems) to another
[analyzing logical statements). The result is
a sort of mental gymnastics for the working
memory complex, as tasks and algorithms
have to be switched rapidly. Not surpris¬
ingly, the WPT is a reasonable indicator
of general intelligence. As it takes only
twelve minutes, it is a cost-effective screen¬
ing instrument.
Philip Roth, a professor at Clemson Uni¬
versity, and his colleagues have conducted a
meta- analysis of the WPT and some other
data on applicant test scores.145 When they
combined the results from all jobs they
found a White-Black d value of .99, and a
White-Hispanic d of .83. Within jobs their
estimate of the White-Black difference was
.83 for jobs of low complexity, .72 for jobs of
medium complexity, and .63 for jobs of high
145 Roth et al.; 2001.
complexity. The reduced d values could be
expected on the basis of self-selection. It was
not possible to analyze the White-Hispanic
difference by job complexity, due to lack of
data.
Meta-analyses have also been conducted
on performance ratings.146 The advantage in
job performance, d^ . 3 in favor of Whites, is
substantially less than the White advantage
on test performance, d^ 1. Black- White dif¬
ferences in job performance are larger when
performance is evaluated by objective cri¬
teria, such as work samples and tests of job
knowledge, than when supervisor ratings are
used. The d values for such measures are in
the .4-. 5 range. This compares to d values
for ratings of interpersonal skills, where the
differences fall closer to .2. 147
There is no contradiction between the
findings on test performance and job perfor¬
mance. Within the population of applicants
for a particular job differences in test scores
are in the d = .y-.8 range. Predictive valid¬
ity correlations range from .3 for jobs of low
cognitive complexity, to .6 for jobs of high
complexity. Multiplying .75 by either .6 or
.3 provides a predicted d of either .45 or .22,
depending upon the cognitive loading of the
job and the performance rating.
Another way to estimate workplace suc¬
cess is to look at income. Income is impor¬
tant in itself, but it is only loosely tied to per¬
formance within an occupation, for many
other variables, such as the type of position,
noneconomic rewards, seniority, and area of
the country also influence income.
White incomes are generally higher than
the incomes of African Americans and Lati¬
nos. Based on their study of the NLSY79
data, Herrnstein and Murray claimed that
the disparity between Black and White
incomes virtually disappears after statisti¬
cally controlling for intelligence, as indexed
by AFQT scores.148 However, they did
not report separate analyses for men and
women. A more detailed analysis of the
146 McKay & McDaniel, 2006; Roth, Huffcut, & Bobko,
2003.
147 McKay & McDaniel, 2006.
148 Herrnstein & Murray, 1994, pp. 324-325.
THE DEMOGRAPHY OF INTELLIGENCE
421
NLSY79 data found that the relation
between income and AFQT scores was
essentially identical across the three racial
groups - Whites, Blacks, and Hispanics -
after making separate analyses for men and
women. The same analyses indicate that the
variation in income associated with AFQT
scores was much smaller than Herrnstein
and Murray had claimed that it was.149
In summary, it appears that there are gaps
in job performance between Whites, African
Americans, and Latinos (to the extent data
is available), and that these gaps are about
what would be predicted from the gaps
in cognitive test scores and the correlation
between test scores and performance.
11.4.7. The Case of Asian- Americans
Asians, a minority in America, but a plu¬
rality in the world, present a very differ¬
ent picture. Asian-Americans, as a group,
have slightly higher cognitive test scores
than do Whites. This is true both today
and at a much earlier time, when Asian-
Americans tended to have lower SES than
they do today. They also have a pronounced
pattern to their differences, on both intel¬
ligence tests and markers of educational
achievement.
The fact that Asian-Americans tend to
have high scores on intelligence tests is not
a new observation. In the 1920-30 period
Stanley Porteus (1883-1972), a professor of
Psychology at the University of Hawaii,
collected data showing that Japanese-
Americans had higher intelligence test
scores than did other groups.150 At that time
Japanese- Americans in Hawaii were consid¬
erably lower on the SES scale than they are
today, so the relatively high test scores rep¬
resented a reversal of the usually observed
positive relation between IQ scores
and SES.
The statement that Asian-Americans
have higher average scores masks an impor¬
tant difference in the pattern of scores.
149 Cawley et al, 1997, especially Table 8.5, and Cav-
allo, el-Abadi, & Heeb, 1998, Table 8.6. See also Fig¬
ure 10.7 in this volume and the accompanying text.
150 Porteus, Dewey, & Bernreuter, 1930.
Asian-Americans tend to do better than
Whites in Mathematics and visual-spatial
reasoning, and tend to be somewhat below
Whites in verbal scores. Figure 11.27 illus¬
trates a typical finding, scores on the SAT for
the period 1997-2007. Black scores are shown
for comparison. The White- Asian difference
is clear, and is also clearly much smaller than
the Black- White difference.151
The US results are mirrored on the inter¬
national scene. Of the forty-eight coun¬
tries participating in the International Math¬
ematics and Science Studies (TIMSS) in
2007, the top five countries were, in order
of eighth-grade mathematics scores; China
(Taipei), South Korea, Singapore, China
(Hong Kong), and Japan. All five countries,
and only those five countries, had test scores
that were reliably higher than the scores for
the United States. The PISA assessments
show a similar picture. In the 2007 PISA
mathematics assessments three out of the
top five countries - Japan, South Korea, and
Hong-Kong (People's Republic of China) -
were from northeast Asia. While northeast
Asian countries generally did well in the
reading assessments, they were not nearly
as dominant as they were in mathematics.
Only Korea retained a position in the top
five.152
These remarks refer strictly to northeast
Asians - people who emigrated from, or
whose forefathers emigrated from, China
(including Taiwan), Japan, or Korea. The
recent emigrations from Southeast Asian
151 People who take the SAT are a self-selected pop¬
ulation, so this comparison might be contaminated
by recruitment effects, as discussed in section 11.1
of this chapter. In this case, however, recruitment
probably works to underestimate the differences
in the population, because the probability that an
Asian-American or Black student will attempt to
take the SAT is higher than the probability that a
White student with equivalent high school exam¬
ination scores will take the SAT. (Calculations by
Tara Madhyastha and the author, based upon com¬
parisons of the frequencies of various ethnic groups
at different NAEP scoring levels and in the self-
selected group taking the SAT.)
152 Hong Kong was tenth and Japan was tied with
the tiny Duchy of Lichtenstein for the thir¬
teenth position. Source: National Center for Educa¬
tional Statistics, Digest of Educational Statistics 2007,
Table 389.
422
HUMAN INTELLIGENCE
♦ White M
•®- Asian M
— A- Black M
• White R
■ Asian R
— Black R
Figure 11.27. SAT scores for three racial/ethnic groups for the
period 1997-2007. The top panel shows scores on the SAT-M (M
code], the bottom shows the SAT-R (R code] scores. Source:
College Board.
nations, such as Thailand, India, Indonesia,
and the Philippines, are not reflected in most
of the statistical comparisons. In fact, the
current US census and educational practice
of lumping all the Asian countries together,
and further adding in emigrants from the
Pacific Islands, is bound to confuse the statis¬
tical summaries. This is going to be a serious
problem in the future, for there are studies
indicating that educational performance and
relevant social conditions vary considerably
across different Asian groups.
In summary, there is little doubt that IQ
scores and educational data present a con¬
sistent ordering of the major racial/ethnic
groups in the United States, with (north¬
east] Asians slightly ahead of Whites, His-
panics about .5 to .7 deviation units below
Whites, and African Americans from .8
to 1.1 deviation units below Whites. The
picture has changed over time, but slowly.
Two questions remain to be addressed; does
this ordering matter, and if it does, what are
its causes?
11.4.8. Are the Tests Valid for Various
Racial and Ethnic Groups within the
Developed Societies?
There have been numerous claims that cog¬
nitive tests do not fairly assess minorities.
In one of the most famous of these dis¬
putes, Larry P. vs. Ryles, 153 a federal court
forbade the use of intelligence tests to clas¬
sify students as “educable mental retardates”
and then channel them into special classes.
(Ironically, eighty years earlier the French
153 Larry P. v. Riles, U.S. Court of Appeals, 1984,
793F2CI969 (9th Circuit].
THE DEMOGRAPHY OF INTELLIGENCE
4*3
Ministry of Education had commissioned
Binet to develop tests for this purpose.)
The courts’ argument was that the tests
were illegally discriminatory because when
they were used a disproportionate percent¬
age of African American children received
IQ scores below 70, one of the criteria for
retardation.
The reasoning used to win the Larry P.
case is typical of the arguments made against
relying on the various cognitive tests as valid
indicators of the mental abilities of minor¬
ity group members. Under what conditions
might the argument be valid?
Return to the distinctions among three
sets of cognitive skills; assessed relevant skills ,
those used both to take the test and to per¬
form in school or the workforce; assessed
irrelevant skills that can be used to take the
test but are irrelevant to school and work¬
place performance; and unassessed relevant
skills, cognitive (or other) skills that are rele¬
vant in the school and workplace, but are not
evaluated by the tests. If only assessed rele¬
vant skills existed, the correlation between
test scores and criterion performance in the
school or workplace would approach 1.0.
This, of course, is an unrealistic criterion,
for no one thinks either that the tests are
perfect or that success is determined solely
by cognitive characteristics. An argument
against testing different racial/ethnic groups
cannot be made on the basis of imperfect
testing.
The argument that the tests are not fair
(to African Americans, Hispanics, or any
other group) cannot rest on differential dis¬
tributions of assessed relevant skills in the
White and minority communities, because
any prediction based only on these skills is
a fair prediction. The argument has to be
made on the basis of assessed irrelevant skills
or unassessed relevant skills. Two conditions
are possible:
a. the members of the affected minority
group are less likely to possess assessed
irrelevant skills than Whites are, or
b. the affected minority group members
are more likely to possess unassessed rel¬
evant skills than Whites are.
If either of these conditions hold the tests are
unfair, either because whites receive a boost
due to their narrow test-taking skills, condi¬
tion (a), or because minority group members
do not receive appropriate credit for their
unassessed relevant skills, condition (b).
Suppose condition (a) exists. Minority
group members will have spuriously lower
test scores than Whites because the White
scores are elevated by their assessed irrel¬
evant skills. The correlation between test
scores and criterion performance, calcu¬
lated only for white examinees, should be
lower than the same correlation calculated
within the minority group, because the
White scores are affected by factors irrel¬
evant to school and workplace (criterion)
performance, while the minority scores are
not. If a common prediction equation is
developed for all examinees, regardless of
race or ethnic status, the equation should
overpredict White performance because the
White scores are systematically and erro¬
neously inflated. Minority performance will
be underpredicted.
Suppose condition (b) exists: minori¬
ties exceed Whites in the possession of
unassessed relevant skills. The correlation
between test scores and criterion perfor¬
mance will be lower in the minority group
than in Whites because the minority cri¬
terion performance scores have a source
of variance that the White scores do not.
If a common prediction equation is used,
minority group performance will again be
underpredicted, due to the minority group
members’ having unassessed but relevant
cognitive skills.
In either case of unfairness, minority
group scores should be underpredicted by an
equation based on scores from all groups -
the common prediction equation. Depend¬
ing upon which case of unfairness occurs,
the correlation between test and criterion
performance may be either higher or lower
when calculated for minority group mem¬
bers alone than for Whites alone.
This definition of fairness is restricted in
two ways. The prediction equation must act
only on test scores. If the prediction equa¬
tion includes an adjustment for racial/ethnic
HUMAN INTELLIGENCE
424
status, by adding or subtracting a con¬
stant, predictions, but not correlations, will
be adjusted appropriately. Whether or not
such an adjustment should be made is a
legal/ethical question rather than a scientific
one, and will not be further discussed.
Psyohometric fairness has been defined
entirely on the basis of statistical accuracy
and skill possession. It is irrelevant to larger
issues of social justice. It could be that when
test scores are used to predict performance
the appropriate equation systematically pre¬
dicts lower criterion scores for minority
group members than for Whites, and that
in fact they have lower criterion scores,
because they have not had the same oppor¬
tunities Whites have had to acquire the
necessary assessed, criterion-relevant skills.
Such a situation is indeed unfair, in a global
sense, and some form of affirmative action
may be appropriate. Blaming an IQ or other
cognitive test for this situation amounts to
shooting the messenger. A test can only
assess skill possessed at the time of testing;
it cannot tell us how those skills came to be
possessed.
How do the facts stand up to the criteria
of fairness?
There is scant evidence that educational
assessment tests are unfair, in the sense
just defined. The validation data for the
SAT provide a clear case. The correlation
between SAT scores and first-year grade
point average, the usual criterion for such
studies, is .53 for Whites, .50 for Hispan-
ics, and .47 for Blacks. While these small
differences are statistically significant (with
sample sizes of 10,000 or more per group),
it is hard to imagine a situation where
they would be of practical importance.
What is even more telling is that African
American and Hispanic grades in college
are systematically overpredicted. This is
not consistent with either condition of
unfairness.154
Differences between White and minor¬
ity group academic performance (White-
minority group) are on the order of d = .40
154 Mattern et al., 2008, Table 2. Correlations have been
corrected for restriction in range.
for African Americans and .30 for Hispanics.
These figures are substantially smaller than
the corresponding differences in test perfor¬
mance, d = 1.0 and d = .6 -.7. 155 This is
what one would expect, because the tests
are imperfect predictors. Assuming a .5 pre¬
dictive validity for educational tests such as
the SAT, the expected d values for academic
performance are .50 and .30-35, not far from
the observed differences.
I have found relatively few industrial
studies that have reported a contrast
between the predictive accuracy of tests
for Whites and minority group members.
A 2008 discussion of this issue by a knowl¬
edgeable researcher in the field156 cited a
study almost twenty years earlier as evidence
that there is no difference in the predictive
accuracy of industrial tests across racial and
ethnic groups. The study in question157 was
a National Academy of Science review of
yet earlier studies, involving a single test,
the Department of Labor’s General Apti¬
tude Test Battery (GATB). Its authors con¬
cluded that test scores were equally predic¬
tive for both groups.158 In my own searches
I found one (1) recent study, of white-
collar/managerial performance in a govern¬
ment agency. It concluded that various
predictors of job success, including but not
limited to cognitive tests, behaved in the
same way for Black and White employees.159
In sum, the tests are accurate predic¬
tors of achievement for the three major
racial/ethnic groups in the United States -
Whites, African Americans, and Hispanics.
The data for Asian- Americans is too scanty
to make a firm statement, although there
is some evidence that academic scores are
underpredicted. Since test scores are higher
155 Sackett, Bomeman, & Connely, 2008.
156 Ibid.
157 Hartigan 8t Wigdor, 1989.
158 Many investigators must have had access to data
relevant to this issue. For example, military studies
have been reported in which it would have been
impossible not to have obtained data that could be
used to compare predictions across groups. Consid¬
ering the amount of controversy over the question
of test bias, it is surprising that the comparisons
were not made.
159 Pulakos, Schmitt, & Chan, 1996.
THE DEMOGRAPHY OF INTELLIGENCE
for Asian-Americans than for Whites, this
suggests that unfairness condition fb) holds:
Asian-Americans may have more unassessed
relevant cognitive skills than do Whites.
The differences in test scores across
racial/ethnic groups almost certainly reflect
a real difference in the distribution of cog¬
nitive skills across racial/ethnic lines. What
might cause the difference?
11.4.9. Social Causes for Disparities in
the Distribution of Intelligence across
Racial/Ethnic Lines
Neither I nor anyone else knows the cause
of the differences in indices of intelligence
among various racial and ethnic groups. Fur¬
thermore, there almost certainly is not any
single cause, and the causes may vary for dif¬
ferent comparisons. We do have some leads.
Not surprisingly, the explanations fall
into two broad categories - social and biolog¬
ical. The social ones will be discussed here,
the biological ones in the next section.
EFFECTS THAT MAY ALTER THE
INTERPRETATION OF TEST SCORES
IN AFFECTED GROUPS
There have been a number of claims that the
tests used for personnel screening are unfair
because they contain questions that draw
on cultural information not available, or less
available, to Blacks, Hispanics, and Asians.
While it is not generally acknowledged, this
claim has to be made with the admittance
that the relevant cultural information has
predictive value for performance, for other¬
wise it would amount to unfairness condi¬
tion [a], as defined in the previous section,
and we have seen that condition (a} cannot
be maintained. The argument has to be the
"social justice” argument that the minority
group members are less able to prepare for
the tests than Whites are.
Arthur Jensen is one of the strongest
critics of this claim. 160 He has applied the
method of correlated vectors, with g load¬
ings replaced by expert judgment of the
160 This is discussed extensively in Jensen’s 1980 book
on the topic. Further discussions of more recent data
can be found in Jensen, 1998, 359-367.
425
extent to which questions or tests are cul¬
turally loaded, and concluded that the cul¬
tural loadings do not predict the extent
of group differences on tests, subtests, and
individual questions, while the g loadings
do. Most of the studies Jensen considered
compared Whites to African Americans in
the United States. A substantial European
study by Michelle Helms-Lorenz and her
colleagues at the University of Tilburg, in
the Netherlands/61 which also used the
method of correlated vectors, found that
cultural loadings rather than g loadings were
the best predictors of differences between
racial/ethnic groups of Dutch children.
The Dutch and the American studies do
not necessarily conflict, for different ethnic
groups were involved. In particular, many of
the African immigrants to the Netherlands
are relatively recent arrivals, while African
Americans in the United States, with a few
exceptions, have a history that goes back
centuries.
In addition to conducting analyses of
racial/ethnic differences on battery type
tests, Jensen has pointed out that White-
Black differences are found on tests that
do not have obviously large cultural biases.
For example, progressive matrix tests do not
have obvious culture loadings, yet they pro¬
duce fairly large racial/ethnic differences.
The matter is somewhat up in the air.
However, the indeterminancy tells us some¬
thing. If cultural unfairness were a major
cause of racial/ethnic differences in test per¬
formance, we would not have as much trou¬
ble detecting it as seems to be the case. This
does not mean that culture plays no part in
determining test scores, particularly if the
cultural differences are marked. For exam¬
ple, I would be suspicious of the use of the
tests to evaluate intelligence in recent immi¬
grants to a developed nation, although test
scores might be useful as an index of assim¬
ilation. Whether it was desirable to use a
valid, but in one sense “culturally unfair,”
test to screen applicants for jobs or academic
programs would be a policy issue, not a sci¬
entific question.
161 Helms-Lorenz, van de Vijver, & Poortinga, 2003.
426
HUMAN INTELLIGENCE
This point was illustrated in an analysis
of several studies of the workplace perfor¬
mance of various immigrant groups in the
Netherlands.162 Immigrants did indeed have
lower test scores than ethnic Dutch. They
also had lower scores in academic achieve¬
ment and on a variety of workplace per¬
formance ratings. (Note the similarity to
the American studies discussed earlier.) The
authors then contrasted test scores and aca¬
demic and workplace performance of suc¬
cessive generations of immigrants. The dif¬
ference between the Dutch and non-Dutch
groups was reduced in the second genera¬
tion. As the immigrant-derived groups had
more experience with the Dutch society
they acquired the cognitive skills it valued.
The test scores responded to the presence of
these skills, and so did academic and work¬
place performance. There was a socially
induced change in intelligence.
This brings us to a slightly different view
of cultural bias on a measure of intelli¬
gence. If one is trying to understand or pre¬
dict success in a certain society - usually a
post-industrial society - then one’s concern
should be with the measurement of intelli¬
gence as defined by that society. Therefore,
a certain amount of cultural bias in a test is
appropriate. How much depends upon the
situation. Troubles occur only if an inves¬
tigator tries to define intelligence as some¬
thing that is invariant across cultures.
A second argument for the idea that test
scores are biased against minorities is that
the testing situation itself is biased. Some
years ago it was claimed that minorities,
and especially minority children, would not
communicate adequately with White exam¬
iners. Very little evidence has surfaced to
support this idea, and in any case it would
not apply strongly to a group testing situ¬
ation. A more serious concern has to do
with motivation. The idea is illustrated by
the stereotype threat argument - that if
minorities (or women) perceive a test as
the sort of test on which they should do
poorly, they will not put as much effort into
162 Te Nijenhuis et al., 2004.
solving test questions. 163 The argument here
is exactly the same as the argument pre¬
sented to explain the fact that women tend
to have lower scores than men on mathe¬
matics and visual-spatial tests. And the evi¬
dence is the same. Stereotype threat can
be demonstrated in laboratory settings/64
but not in field situations where success
or failure has real consequences for the
examinees.165 The use of the MGCFA tech¬
nique to evaluate stereotype threat effects in
high-stakes testing has been suggested, but
at present no such studies are available.166
A third argument is that minority group
members may lack culturally relevant cogni¬
tive skills that are evaluated by conventional
tests, but that they possess other important
cognitive skills that are not evaluated by the
tests. This is condition (b) in my categoriza¬
tion of unf air testing. It implies that the tests
should underpredict minority performance.
Such underprediction has not been reported
for African Americans or Hispanics, either
in academia or the workplace. There is a
suggestion that condition (b) may apply to
some Asian- Americans.
The claim that test scores are irrelevant or
inaccurate in predicting minority group per¬
formance is contradicted by the evidence.
SOCIAL EXPLANATIONS: DEVELOPMENT
Some explanations for racial and ethnic dif¬
ferences in intelligence accept the validity of
test scores as indicators of actual differences,
and then seek to explain these differences
in terms of social causes acting on cognitive
development. In order to understand these
explanations it helps to step back and con¬
sider how we develop intelligence.
Intelligence refers to the possession of a
set of cognitive skills and knowledge that are
useful in post-industrial society. These skills
are not inherited directly, in the sense that
eye color is inherited. They are produced by
an interaction between a person’s learning
163 Steele & Aaronson, 1995, 1998; Suzuki & Aaronson,
2005.
164 Nguyen & Ryan, 2008.
165 Cullen, Hardison, & Sackett, 2004; Sackett, Hardi¬
son, & Cullen, 2004; Strickler & Ward, 2004.
i66Wicherts, Dolan, & Hessen, 2005.
THE DEMOGRAPHY OF INTELLIGENCE
427
capacity, which is initially genetically deter¬
mined but soon influenced by what the per¬
son already knows, and social supports that
facilitate the learning process. Initially these
social supports are largely within-family, but
as a person grows older the support net¬
work ranges from kindergarten to the uni¬
versity, church, and workplace, and from
exploration of the playground to exploration
of the World Wide Web. Cognitive skills are
required to utilize these resources, so there is
an important positive feedback loop. What
you know already is a powerful predictor of
what you are about to learn.
Human learning is an active process. It
depends upon both the potential offered by
the environment and the student’s motiva¬
tion and skill in interacting with the environ¬
ment in a way that furthers the acquisition
of learning. These statements apply equally
to learning calculus, learning how to cook a
souffle, and learning how to hit a backhand
overhead shot in tennis.
The next step is to apply these statements
to racial/ethnic differences in intelligence
and academic achievement.
Parental socioeconomic status (SES) is
commonly pointed to as an indicator of
social support [or lack thereof). Parental
SES is confounded with racial/ethnic sta¬
tus. This is particularly true for recently
arrived immigrants, who also may not be
familiar with the dominant language. Chap¬
ter 9 reviewed the substantial evidence, from
cultures as diverse as the rural Philippines
and the American Midwest, showing that
children from families that are disrupted
and/or socioeconomically stressed are read
to less, are less encouraged to do problem
solving on their own, and generally tend to
have lower intelligence test scores, even in
the pre-school years. Home environments
count.
Because home environment is strongly
correlated with SES, one could argue that
to a great extent the observed racial/ethnic
disparities in cognitive skills are actu¬
ally SES effects. This proposal has to be
taken seriously, for the correlation between
race/ethnicity and SES is substantial. In 2006
the percentages of families with children
under eighteen who were below the poverty
line was as follows:
White . . . 9.5%
Black . . . 33.0%
Hispanic . . . 26.6%
Asian . . . 12.0%
A lack of financial resources substantially
constrains the amount of time and effort that
adults can give to parenting. This undoubt¬
edly takes its toll on the children. Recall that
the gaps in cognitive achievement toward
the end of high school can be predicted, on
an individual basis, by the cognitive gap at
the start of elementary school. The original
gap is closely related to SES.
Nevertheless, there are three reasons for
not accepting SES as the sole explanation
for racial/ethnic disparities in intelligence.
The first is statistical: SES measurements do
not fully account for the differences in intel¬
ligence measures between groups. An SES
argument also fails to account for the fact
that the differences between groups on cog¬
nitive measures tend to be large for “cultur¬
ally reduced” tests, such as the RPM tests.
Second, SES is itself confounded with
intelligence. We then get a chicken- and-egg
phenomenon. For children, parental SES is
confounded with parental intelligence, and
since intelligence is heritable to a significant
degree, it is not clear whether an effect of
parental SES on children's intelligence is due
to social or biological inheritance.
Finally, and to me most compelling,
socioeconomic status is a statistical abstrac¬
tion. As such, it cannot cause anything.
Intellectual development depends upon
physical and social variables within the envi¬
ronment that are correlated with SES.
What might some of these variables be?
One class of variables has already been
implicated: parental practices. It has been
estimated that in order to participate fully
in school a child has to have learned to
read approximately 9,000 words by the third
grade. The child also has to have a good
knowledge of the syntactical and semantic
rules of the language that will be used in the
school. A substantial portion of this knowl¬
edge is acquired via parental interaction,
428
HUMAN INTELLIGENCE
such as reading to children/67 While there
is less data for other skills, such as encourag¬
ing problem solving (at least that I am aware
of); the same thing is likely to be true of ele¬
mentary mathematics (e.g., counting] and
general problem-solving skills.
Parenting practices are required to
draw out biological potentials in young
children. l6R The extent to which these are
developed at the start of school is highly
predictive of later development in school.
As a general statement, for there obviously
are exceptions, Black, White and Latino par¬
ents with low SES seem to be low on the
time-consuming parental skills required to
guide children toward independent problem
solving. Asian parents with similar low SES
give the practice of these skills high prior¬
ity. This seems to be particularly true of
Asian- Americans from the northeast Asian
nations - China, Korea, and Japan/69
If this argument is correct, it ought
to be possible to account for the gap in
Black-White performance by considering
the effect of these variables. This question
was addressed in an analysis of data from the
Panel Study on Income Dynamics (PSID], a
longitudinal study being conducted by the
Institute for Social Research of the Univer¬
sity of Michigan. In 1997 extensive data was
collected on the families involved who had
dependent children age twelve and under.
The children's cognitive skills were mea¬
sured using age-appropriate tests, including
several of the tests in the Woodcock-
Johnson battery. Parental income and occu¬
pation and household wealth were mea¬
sured. Mother’s verbal skills were measured,
using the paragraph comprehension subtest
of the Woodcock-Johnson battery. This is a
good marker for Gc. In addition, a number
of measurements were made on the home
environment itself, including parenting style
and the quality of adult interactions with
children.
The typical Black- White gap appeared
when the children’s cognitive scores were
167 Wolf, 2007, Chapter 4.
168 Collins et al., 2000.
169 Corwyn & Bradley, 2008; Steinberg, 1996.
considered alone. However, it was greatly
reduced, and no longer statistically reliable,
after the family and home characteristics
were considered. The three best predictors
were occupational status of the head of
the household (the higher prestige of the
occupation, the better the child’s scores),
mother’s score on the verbal comprehen¬
sion test (the higher the mother’s score, the
higher the child’s score), and the measures
of the nature of the interactions with the
child/70
The variables that best predict a child’s
scores are variables correlated with racial
status. This is shown in Figure 11.28, which
presents the d values associated with the
four measures that were found most pre¬
dictive of children’s cognitive test scores.
Black and White families differ on the key
variables. This is an illustration of the gen¬
eral point that socioeconomic characteris¬
tics, rather than racial identity per se, are
determiners of the Black- White gap in fairly
young children.
What about group differences in older
children and adults? The influence of fam¬
ily environments is diminished as children
age. Therefore, we want to look at the
interactions between students in different
racial/ethnic groups and their schools, for
the development of intelligence during this
period depends on three things: biological
potential (which can never be gainsaid), the
quality of instruction, and the way in which
the student interacts with the instruction.
Laurence Steinberg, a professor at Tem¬
ple University, and his colleagues have con¬
ducted extensive studies of the attitudes
of adolescents toward school/71 Steinberg’s
group reached the following conclusions:
1. The dominant attitudes of the White
peer group discourage high levels
of school achievement. Conspicuous
efforts toward studying are frowned
upon. However, poor academic perfor¬
mance is also looked down upon. In
general, it is OK to do well in school,
170 Yeung & Conley, 2008.
171 Steinberg, 1996.
THE DEMOGRAPHY OF INTELLIGENCE
429
Figure 11.28. The effect sizes for White-Black contrasts on selected
variables shown to predict the Black- White contrast on tests of
cognitive skills. All effect sizes are reliable at the .05 level or lower.
Calculations based on Yeung & Conley, 2008, Table 3. The
selection of variables was based on regression analyses presented in
that paper.
provided that you do not appear to be
working hard at it.
2. Black and Hispanic students claim
that high achievement would be valu¬
able, but do not seem to be threat¬
ened by poor achievement. Consider¬
able emphasis is placed upon behaviors
that display group solidarity; “hanging
out” is seen as important, studying is
not.
3. Asian- American students believe in the
importance of high achievement, and
expect that it will be achieved only by
sustained effort. This group was the only
group that reported substantial discus¬
sion of class work between peers. The
effect is highest in children who are
immigrants or whose parents are immi¬
grants from the three northeast Asian
countries.
John Ogbu, an anthropologist at the
University of California, Berkeley, reached
much the same conclusion on the basis of
ethnographic analyses of the attitudes of
Black adolescents from well-to-do homes in
a similar city. Support for Ogbu’s conclusion
was obtained in a survey of student feel¬
ings about high academic achievement. In
general, Black tenth-graders saw social costs
to academic achievement, especially if they
were in a predominantly Black school. Both
Black and White teenagers believed that
being in an honor society would threaten
their chances of being part of the “in” crowd
or being popular with the opposite sex,
but these beliefs were much more common
among Black students.172
This line of research has generated a great
deal of debate. Several social commenta¬
tors and activists, including African Ameri¬
can activists, have accepted the findings and
made efforts to change adolescent attitudes
in the African American communities. Oth¬
ers have argued that the phenomenon is a
result of socioeconomic status, rather than
racial/ethnic attitudes. As a statistical phe¬
nomenon, though, this debate does not mat¬
ter! If Blacks and Latinos are concentrated
in lower SES groups, then any gap in per¬
formance that is based on the performance
of students in general could appear either as
an SES or a racial/ethnic disparity.
Finally, there is the question of the
schools themselves. In extremely segregated
societies, such as the American South prior
to [forced) school integration in the 1950s or
South Africa prior to the end of apartheid,
schooling was separate and grossly unequal
172 Cook & Ludwig, 1998. Perhaps a way to improve
American education is to keep the names of stu¬
dents on high school honor rolls a secret.
HUMAN INTELLIGENCE
430
for Whites compared to non-Whites. That
is no longer the situation today, in either
country, but there is a correlation between
racial/ethnic status, residential location, and
the quality of schooling. For instance, in the
United States schools in heavily Black, low
SES areas tend not to be desirable positions
for many teachers. Also, to the extent that
school financing is dependent upon the local
tax base, schools located in economically
distressed areas are at a disadvantage. This
is especially serious for attempts to control
student/teacher ratios. At the gross level of
measuring school quality by the amount of
financing available per student, school fund¬
ing seems to have little influence on out¬
comes. A finer-grained analysis provides a
different picture.
As was shown earlier, the Black- White
gap in cognitive skills upon finishing school¬
ing is predictable, on an individual basis,
from the gap upon entry to kindergarten.
The size of the gap between groups, mea¬
sured in deviation units, is substantially
larger at the end of high school than at the
beginning of the first grade. This suggests
that the mathematical relationship is
df = cdg + e, (11.3)
where dg is the original black-white gap, df
is the final gap, c is a constant greater than 1,
and e is a residual term. Furthermore, the
equation should hold for any educational
transition.
Equation 11.3 implies that the Black-
White gap grows steadily throughout the
school years, which it does. This could be
taken as evidence for social practices within
the schools that place Black children at
a disadvantage. However, a careful analy¬
sis of the patterns of accomplishment in
both Black and White schoolchildren sug¬
gests a different conclusion. 17^ Equation 11.3
applies not only to Black- White contrasts,
but also to contrasts between White chil¬
dren who enter school in an advantaged or
disadvantaged position. Less abstractly, the
173 Phillips et al, 1998.
gap in performance between students who
were initially well or poorly prepared, prior
to entering kindergarten or the first grade,
increases steadily throughout the school
years. Because Black children are more likely
than White children to enter at a disad¬
vantage, the progressively increasing gap
between well-prepared and poorly prepared
students, regardless of race, can appear as
an increasing gap between Black and White
students.
In sum, there are many social prac¬
tices that have an impact on racial and
ethnic differences in cognitive skills. They
include parenting practices and cultural
beliefs about the educational process. These
in turn interact with issues of school financ¬
ing and teacher behavior. It is extremely
hard to quantify these various social vari¬
ables. Furthermore, they are highly interac¬
tive and collinear, making an assignment of
weights or evaluation of causal models diffi¬
cult. I do not doubt, though, that the social
environment of racial and ethnic groups has
a major influence on the development of
intelligence within those groups.
We might then ask why these differences
in social environments arise in the first place.
Here we enter the realm of speculation.
William Julius Wilson, a sociologist at
the University of Chicago, has argued that
since the end of segregation and legal
discrimination the Black community has
become increasingly bifurcated.174 Middle-
class Black families have seized expanded
opportunities for economic, residential, and
social improvement. To some extent this has
led them to disengagement from the “ghet¬
tos,” heavily Black areas in the inner cities.
At the same time changes in the nature of
work have led to a reduction in the employ¬
ment opportunities available to less edu¬
cated, less well-off people, regardless of their
race or ethnicity. This trend has fallen heav¬
iest on the Black community, because fam¬
ilies there often did not have the financial
and educational backing required to give
their children the opportunity to acquire
174 Wilson, 1997.
THE DEMOGRAPHY OF INTELLIGENCE
431
the credentials that are becoming increas¬
ingly necessary for social advancement. In
Wilson’s terms there is now a permanent
underclass that is not necessarily Black but
tends to be heavily Black.
If we accept Wilson's sociological anal¬
ysis, a psychological corollary follows from
the Challenge Hypothesis stated in Chap¬
ter 1. An underclass that is divorced from
the mainstream of the post-industrial soci¬
ety will fail to develop some of the cognitive
skills that are required in that society. Since
the divorce is not complete, there will be an
overlap between the skills required in both
the underclass society and the more gen¬
eral society. For instance, in modern Amer¬
ica an inner-city Black may not speak the
standard American English dialect, but he or
she can develop the ability to comprehend it
by watching television. The same thing can¬
not be said for the skills involved in math¬
ematical or counterfactual reasoning. It is
worth noting that in support of this explana¬
tion there are studies showing that there are
no differences between Blacks and Whites
in knowledge retrieval, providing that the
opportunity for knowledge acquisition (of
word definitions) is equal.175
Stanley Sue, a professor of Psychology
at the University of California, Los Ange¬
les, has offered a social-belief explanation
for the predominance of Asian-Americans
in both high test scores and academic and
professional achievement.176 Sue rejects the
statement sometimes made that Asians in
America are a modern example of successful
migration. He agrees that they have indeed
succeeded in some fields, but argues that a
careful analysis of mental health indictors
indicates that the success has been accom¬
panied by a good deal of stress. He notes
that Asian-American success has come pri¬
marily in fields such as engineering and
medicine, where education and certification
play a dominant role. Asian-Americans are
less prominent in fields such as law and
175 Fagan & Holland, 2002, 2007.
176 See Sue and Okazaki, 1990, for a good discussion of
Sue’s ideas.
politics, where personal acceptance be¬
comes a central issue.177 Sue argues that
a concentration on scientific and technical
careers, combined with a traditional belief
that hard work will lead to success, can
account for much of the data on Asian-
American cognitive accomplishments.
Ogbu presented an analysis some¬
what similar to Sue’s, but for African
Americans.178 Like Sue, Ogbu emphasized
the importance of historically developed
beliefs within the affected group as deter¬
minants of current behaviors. Ogbu then
went on to distinguish between groups such
as Asian-Americans, who were voluntary
emigrants to the United States, and who
arrived there with a historic belief in the effi¬
cacy of hard work, and African Americans,
who were the descendants of an involun¬
tary migration, and who came from a social
group that, after several generations of slav¬
ery and discrimination, had not been given
very much reason to believe that hard work
would be followed by rewards.
I do not know of any similar anal¬
ysis of the relations between cognitive
achievement, current practices, and histor¬
ical causes in the Hispanic community. In
fact, I doubt that there can be such an anal¬
ysis for the community as a whole, for it
is composed of very different subcommu¬
nities. Puerto Ricans recently from Puerto
Rico, Puerto Ricans who moved to the conti¬
nental US from two to four generations ago,
Cuban immigrants who left Cuba following
the ascension of a communist dictatorship
there in the 1950s, the descendants of Mex¬
icans who moved to present-day California
and the Southwest over two hundred years
ago, and recently arrived immigrants and
177 An exception has to be made for the State of
Hawaii, where Asian-Americans are the major¬
ity ethnic group and, predictably, have dominated
state politics. In the other forty-nine states only
two Asian-Americans have been elected governor
as of January 1, 2010. Both have served recently:
Gary Locke [Chinese descent) in Washington State
(1997-2005) and Bobby Jindal (Indian descent),
elected Governor of Louisiana beginning in January
2008.
178 Ogbu, 2003.
HUMAN INTELLIGENCE
432
children of recent immigrants from Latin
America are very different people. Racial
and ethnic designations are fuzzy concepts.
In the case of American Hispanics the con¬
cept is very fuzzy indeed.
To sum up, there are powerful argu¬
ments and a good deal of evidence for the
proposition that racial/ethnic differences in
intelligence are influenced by a variety of
social variables. There is no evidence point¬
ing to a single explanatory variable, nor is
there likely to be. Statistically, the variables
are highly collinear, so that different causal
models can be said to describe the data to
very nearly the same degree of accuracy. In
many cases what appears to be an impor¬
tant variable turns out to be a proxy for
other variables. For instance, income and
family wealth are statistically predictive of
the Black-White gap, but when noneco¬
nomic family characteristics, such as parent¬
ing style, are statistically controlled for, the
wealth-IQ relation disappears.179 Logically,
there are all sorts of reasons to expect inter¬
actions between variables. Trying to sort out
the relationship between cognitive perfor¬
mance and contemporary indices of noncog-
nitive behaviors is a data analyst’s night¬
mare.
Social commentators disagree about how
the present situation came to pass. Ogbu and
Wilson disagree over whether the situation
for African Americans has developed over
two or three centuries or is due to devel¬
opments in the last half-century. Sue falls
somewhere in between, referring to beliefs
held by an immigrant community that has
been developing from successive waves of
entrants for the last 150 years, but refer¬
ring (somewhat generally) to social tradi¬
tions that have a history of over a thousand
years.
I do not think we are likely to resolve
these issues. As is the case for evolution¬
ary psychology’s "man the hunter” hypoth¬
esis, the stories about the historic origins
of present-day behavioral practices are rea¬
sonable, but to distinguish among them we
would have to have access to historical
179 Yeung & Conley, 2008.
data that has probably disappeared forever.
More generally, rigorous methods of scien¬
tific analysis are difficult to apply to historic
events. A social commentator may be able
to convince us that his or her way of looking
at the origins of racial/ ethnic differences in
intelligence is useful, but is unlikely to ever
have either a sufficiently precise theory or
sufficiently rich data to discriminate among
different viewpoints.
11.4.10. Biological Causes for Racial
and Ethnic Differences
Potential biological causes of racial/ethnic
differences in intelligence may be either
environmental or genetic.
Chapter 9 presented evidence on three
of the most toxic environmental pathogens:
exposure to lead, excessive use of alcohol
and other psychoactive agents (alcohol is by
far the biggest problem), and severe mal¬
nutrition. What data there is indicates that
atmospheric and environmental lead expo¬
sure may be greater in the African Ameri¬
can and Hispanic communities than in the
White community. I know of no data for
Asian communities, and in this case a dis¬
tinction would have to be made between
recent immigrant groups and others.
Alcohol abuse is more prevalent in the
Black and Hispanic groups than in the
White and Asian groups. Alcohol abuse is
highly correlated with SES, thus confus¬
ing the issue. Failure to follow good dietary
regimes is also a problem within the African
American and Hispanic communities and,
like alcohol abuse, is correlated with SES.
Whether nutritional deficiencies in the two
minority communities are severe enough to
be a significant threat to cognitive develop¬
ment is debatable, although there certainly
are international examples of severe malnu¬
trition in Africa and Latin America.
What we lack are quantitative stud¬
ies determining whether differential expo¬
sures to biologically hazardous environ¬
ments (including self-induced ones) could
account for the substantial gap in IQ scores
between Whites and the two minority com¬
munities. The situation is not analogous
THE DEMOGRAPHY OF INTELLIGENCE
433
to the problem of discriminating between
alternative accounts of history. Questions
about the role that environmental hazards
play in determining racial/ethnic differences
in intelligence can be answered. As far as I
know, though, the needed research has not
been done.
There is a great deal of contention
over the role that genetic differences play
in establishing racial/ethnic differences in
intelligence. The following two quotes
summarize the conclusions of two recent
reviewers:
For the race differences in IQ, we can be
confident that genes play no role at all.
Nisbett, 2009, p. 197
. . . genetic and cultural factors carry the
exact same weight in causing the mean
Black-White difference in IQ as they do
in causing individual differences in IQ,
about 80% genetic-20% environmental by
adulthood.
Rushston <Sd Jensen, 2005, p. 27Q
There seems to be a difference of opinion.
The argument for genetic causes for
group differences has been maintained by
several serious researchers over the years.
The three most prominent advocates of this
position today are Arthur Jensen of the
University of California, Berkeley; Richard
Lynn of the University of Ulster; and J.
Phillipe Rushton of the University of West¬
ern Ontario. The arguments they propose,
which are essentially identical, were well
presented in a 2005 paper by Rushton and
Jensen.18 Before discussing these claims,
a comment about how genetic effects
might influence racial/ethnic differences is
in order.
No one inherits a test score. In order to
make a direct argument for a genetic cause
for group differences in intelligence one has
to show that there is some biological char¬
acteristic that can account for differences in
intelligence between the groups being dis¬
cussed and that the characteristic is known
or reasonably believed to be under genetic
180 Rushton & Jensen, 2005.
control. Two such characteristics have been
proposed.
One is information-processing capacity,
and especially speed of information pro¬
cessing, as measured by elementary cogni¬
tive tasks such as reaction time paradigms.
Rushton and Jensen argue that there is a
substantial difference in speed of process¬
ing between Whites and Blacks, in favor
of Whites, and smaller differences between
Whites and Asians, in favor of Asians.
The evidence they offer is based largely
on Jensen's studies of differences in reac¬
tion times between Whites and African
Americans on fairly simple tasks.181 How¬
ever, a review of a wider range of stud¬
ies of racial/ethnic differences in reaction
time and other information-processing tasks
found considerable inconsistency in results.
The reviewers concluded that the difference
between Blacks and Whites, which would be
expected to be the largest difference, was on
the order of d <.20. 182 Given that the correla¬
tion between speed-of-processing measures
and IQ scores is around .5, after correction
for various artifacts, the expected contribu¬
tion of speed-of-processing measures to the
Black- White test score gap would be .10,
while the gap itself is on the order of 1.0.
A lot remains to be explained.
Rushton and Jensen, and (in separate
papers) Lynn have also proposed that the
difference between groups in test scores
is due to differences in brain size. Brain
size does have a correlation of about .35
with intelligence within the White popula¬
tion. Brain size is almost entirely genetically
determined.18* Therefore, evidence for sub¬
stantial differences between racial/ethnic
groups in brain size would be an impor¬
tant link in an argument for a genetic
basis for group differences in intelligence.
However, such studies would be difficult
181 Jensen, 1993. This study can be criticized for having
given far fewer practice trials, fifteen, than is usual in
reaction time studies. Studies within cognitive psy¬
chology have shown that speed measures become
limiting factors in reaction time studies only after
dozens, if not hundreds, of trials. See Ackerman,
1986, 1987, for a discussion.
182 Shepard & Vernon, 2008.
183 Baare et al., 2001.
434
HUMAN INTELLIGENCE
to arrange, due to the expense of obtaining
brain images. Therefore, researchers inter¬
ested in this topic have made estimates of
brain size differences from external mea¬
sures on the skull. This indirect method has
its problems.
The correlation between intelligence and
cranial capacity, estimated from measure¬
ments on the skull, drops to about .2, which
is not surprising as brain size is substan¬
tially but imperfectly related to skull size. In
studies by Rushton the difference between
White adults and Black adults in cranial
capacity is 43 cm3, which corresponds to a d
for cranial capacity of .46.184 Combined with
a .2 correlation, this leads to the conclusion
that on the basis of skull size there should
be a difference of d = .09 between Black and
White test scores. If we accept the idea that
brain size and processing speed are statisti¬
cally independent, the expected gap due to
these factors is then .19, still far below the
observed value of 1.0.
Richard Lynn has argued that changes
in intelligence over the past fifty years, the
cohort (Flynn] effect described in Chapter 9,
are probably due in part to changes in
nutrition.185 And he has used the correla¬
tion between cranial capacity and intelli¬
gence test scores to buttress his argument,
for cranial capacity, unlike brain size, does
have a substantial environmental compo¬
nent. To the extent that Lynn is right about
nutrition, this would reduce the genetic
effect upon intelligence associated with
group differences in measurements on the
skull.
As further evidence of a genetic basis for
Black- White differences Rushton and Jensen
point to various studies in which the method
of correlated vectors was used to show that g
loadings on tests correlate with Black- White
184 This is based on the figures provided in Rushton,
1992, p. 405, text and Table 2. Rushton does not
present a d value. I have calculated it by estimating
standard deviations within groups provided in his
Table 1, and then computing the average within-
group variances. The mean difference was taken
from the numbers given in the text, adjusted for
military rank and sex.
185 Lynn, 1998.
differences and with heritability estimates.
The ambiguities in the method of corre¬
lated vectors have already been commented
upon.
Rushton and Jensen also make a good deal
of what they refer to as “admixture stud¬
ies,” in which mixed-race groups are shown
to have intelligence test scores intermediate
between those of Whites and Blacks. In gen¬
eral, such studies are done outside of the
United States, (e.g., in South Africa), where
“Black” and “White” are more clearly defined
groups than they are in the US. The mixed
groups are generally intermediate on a vari¬
ety of social and educational measurements.
The studies are ambiguous unless these vari¬
ables are controlled.
Rushton and Jensen, and in other writ¬
ings (to be discussed in the next section)
Lynn make a few more points, but these
are their major ones. In general, I find their
arguments not so much wrong as vastly over¬
stated. But overstatement does not mean
that there is no point to them. Rushton and
Jensen present two conclusions.
1. They argue that the hypothesis that
Black- White (and, by extension, Asian
and Hispanic) differences in intelli¬
gence are entirely due to environment,
the hypothesis they refer to as the
100% environment hypothesis, cannot
be maintained.
2. They propose as an alternative “default
hypothesis” that the Black- White differ¬
ence is 80% due to genetic differences.
They base this conclusion on the obser¬
vation that within Whites intelligence
test scores have a heritability coefficient
of .8.
Rushton and Jensen (and Lynn) are cor¬
rect in saying that the 100% environmental
hypothesis cannot be maintained. Nisbett’s
extreme statement has virtually no chance
of being true. However, the 100% environ¬
mental hypothesis is something of a stalking
horse. Many researchers who are primarily
interested in environmental differences asso¬
ciated with racial and ethnic differences in
THE DEMOGRAPHY OF INTELLIGENCE
435
intelligence would not be at all perturbed
by an ironclad demonstration that, say, 3% of
the gap is due to genetic differences. The real
issue is over the identity and size of genetic
and environmental influences on group dif¬
ferences in intelligence, not the existence of
either one.
The 80% default hypothesis is an extreme
and excessively precise statement. It is based
on the assumption that the factors that con¬
tribute to the between-group differences
are the same as the factors that contribute
to within-group differences. This is doubt¬
ful. The genetic variability associated with
the "continent of origin” distinction (races,
for short) is much less than the variability
between people with the same continent of
origin. Estimates have placed the between-
group correlation at from to 5% to 15% of the
total variation within the human genome.
On a statistical basis this is quite enough
to identify genetic clusters associated with
racial groupings.186 There are certainly envi¬
ronmental differences between racial/ethnic
groups. Do they amount to 5% of the total
environmental variation in the United States
population? Are the genetic and environ¬
mental variations associated with intelli¬
gence distributed between groups in exactly
the same proportion as they are distributed
within groups? We have no idea.
This is a topic about which there is a
great deal of emotion and an unfortunate
paucity of data. The direct evidence that
we have for genetic effects does not come
close to accounting for the size of the gap
between White and African American test
scores. Neither do environmental effects.
And, unfortunately, the environmental evi¬
dence has often been presented as evidence
that environmental effects do occur — which
no advocate of genetic models has ever
denied - but has not been presented in a
way that permits a quantitative estimate
of how important environmental effects are
in determining group differences in intelli¬
gence in the population.
i86Bamshad et al.( 2004; Edwards, 2003; Tang et al.,
2005.
The questions raised by the existence
of racial/ethnic group differences in intelli¬
gence are complex. The genome exerts both
distal and proximal effects, so sometimes it
is not clear whether a particular influence
should be regarded as a genetic or an envi¬
ronmental one. Consider the case of par¬
enting practices, shown to be an influence
on a child’s development of intelligence.
If these practices are partly under control
of the parents’ genomes, and partly under
the control of parental education, then the
effect on the child could be called either
environmental or genetic! Or an interaction
between the two. Or we could say that the
effect is genetic, but that it can be mod¬
ified by education, just as any number of
genetic disorders can be modified by medical
treatment.
In spite of their complexity, questions
like these are answerable. We can conduct
quantitative searches for proximal and dis¬
tal influences, on both the genetic and envi¬
ronmental sides. While we do not know,
today, what genes are associated with intel¬
ligence, we will find them. When the genes
are identified it will be possible to deter¬
mine allele frequencies across different pop¬
ulations (with the caution that if the groups
are in contact these frequencies will change
over time!). Studies of environmental influ¬
ences could go beyond the demonstration
stage to the stage of making estimates of
the size of these effects outside of the
laboratory - once again remembering that
we will be shooting at a moving target,
for environments drift faster than gene
frequencies.
This summary will probably not satisfy
those who have taken strong stands on either
side of the debate over racial and ethnic
differences in intelligence. Bold hypotheses
“rally the troops” and make great entrees
for television talk shows. People who take
intermediate positions are said to be “wishy-
washy” or “afraid to say what they really
think.” Nevertheless, the issue is complex,
and oversimplifications do not help. There
are group differences in intelligence, they
are important, and there are both scientific
436
HUMAN INTELLIGENCE
and social reasons for trying to understand
them. Plausible cases can be made for both
genetic and environmental contributions
to differences in intelligence. The evidence
required to quantify the relative sizes of
these contributions to group differences is
lacking. The relative sizes of environmental
and genetic influences will vary over time
and place. Some of these influences may
be amenable to change, while others will
be resistant to change. The relevant ques¬
tions can be studied. Denials or overly pre¬
cise statements on either the pro-genetic
or pro-environmental side do not move the
debate forward. They generate heat rather
than light.
And this is what I really believe!
11.5. The International Picture
Wealth and health are not distributed uni¬
formly across nations; the countries of
Europe and North America and their deriva¬
tive oceanic cultures are best-off; Japan,
Korea, and more recently China are closing
the gap; sub-Saharan Africa is worst-off. The
Middle East and South Asian nations and
Latin America fall in between. Neither the
genetic composition nor the socioeconomic
histories of these regions are the same. To
what extent is the present unequal distribu¬
tion of health and wealth due to differences
in the cognitive capacities of different peo¬
ples, and to what extent must these differ¬
ences be taken into account as economies
are developed over the globe?
This is an old question. Herodotus specu¬
lated about different temperaments of peo¬
ples in the fifth century BCE. From the
seventeenth century until modern times
economists have discussed the topic under
the general title human capital. This term
is used collectively to refer to the devel¬
oped capabilities of a workforce, which can
vary from being literate to having a high per¬
centage of people with engineering degrees.
It is generally agreed that human capital is
very important in determining the economic
potential of a state or nation. Political lead¬
ers are more likely to justify taxes to support
schools by claiming that an educated popu¬
lation has a greater economic potential than
an uneducated population than by claiming
that the educated have a moral and philo¬
sophical advantage. l87
Economists seldom use the word “intel¬
ligence.” They are interested in the dis¬
tribution of developed skills in a society,
and the implications of the distribution.
How those skills came to be are of less con¬
cern to them. Economists are also well aware
that for some people the word “intelligence”
has connotations of fixed, genetically deter¬
mined constraints on mental capacity. Any¬
one interested in influencing policies con¬
cerning “human capital” (the economists’
euphemism for intelligence) is well advised
to avoid using the word, for if someone
does suggest that intelligence limits the socio¬
economic development of nations there is
unnecessary Hell to pay. A spectacular case
of this is presented in panel 11.6.
However, some people do look at the
international distribution of intelligence,
and do argue that the present-day (and
presumably past and future) distribution
of wealth, health, and happiness is due
in part to differences in national intelli¬
gence. Richard Lynn, whose work on sex and
race/ethnic differences was discussed ear¬
lier in this chapter, and Tutu Vanhanen, a
Finnish economist, invigorated the field with
the publication of two challenging books
and some related papers. I shall be highly
critical of their empirical work, and even
more so of their interpretations. They do
deserve credit for raising important ques¬
tions in a way that has resulted in interest¬
ing and important findings. Before present¬
ing the Lynn and Vanhanen research and
studies that amplified and redirected it, let
us consider the problems faced by any study
of national intelligence.
187 Interestingly, this is a change in view. Thomas Jef¬
ferson argued for education on the grounds that
it would make people better able to participate in
political, not economic, life. There is some evidence
that intelligence test scores are correlated with the
relatively liberal values that have evolved (substan¬
tially!) from Jefferson's views of government and
society. See Deary, Batty, & Gale, 2008.
THE DEMOGRAPHY OF INTELLIGENCE
457
11.5.1. Methodological Issues
Estimating the distribution of intelligence in
a nation requires a probability sample of the
residents in each nation and the use of tests
that are valid in all the nations under review.
These are stringent requirements. Represen¬
tative national samples are hard to obtain
in the most advanced post-industrial coun¬
tries. The problem is virtually unsolvable in
less developed countries, especially if these
countries are subject to major unrest. As of
2010 there is no way that the governments of
the war-torn countries of the Congo, Soma¬
lia, or Afghanistan could obtain a census or
probability sample of all their residents, for
any purpose.
It is often much easier to approximate
a probability sample in a selected seg¬
ment of the population, such as schoolchil¬
dren (the most likely accessible population)
or, in some countries, registrants for mili¬
tary service. You then face three problems.
The accessible populations may differ across
countries; the estimates will apply only to
the segment of the population involved;
and the recruitment procedures that con¬
struct the accessible samples from the gen¬
eral population may differ across nations.
The problem is particularly acute in the case
of schoolchildren.
Consider the following contrast. In the
United Kingdom essentially all children go
to school, and the government has, from
time to time, given well-standardized cog¬
nitive tests to students. In the Congo large
parts of the country are cut off from gov¬
ernment control, and schooling outside the
major cities is haphazard at best. Inferences
about national intelligence based upon stud¬
ies of schoolchildren in the United Kingdom
and the Congo would hardly be comparable.
These examples are extreme. Differen¬
tial recruitment can occur in well-developed
nations. South Korea reported that 42% of
the participants in the Program for Inter¬
national Students Assessment (PISA) were
girls; France reported 53 percent.188 Any
comparison between the PISA results from
188 Rindermann, 2007.
Korea and France would be confounded by
male-female differences in sample recruit¬
ment.
Obtaining comparable and appropriate
intelligence tests is equally challenging. The
point has already been made that the tests
are weighted toward evaluating skills that
are relevant to the developed world, and that
any test necessarily is partly an evaluation of
the possession of test-taking skills. One has
to be open to the possibility that outside
the developed world people may not have
had a chance to acquire the skills relevant
to industrial and post-industrial society, and
even that the people being examined may
not have acquired the skills required to deal
with the test-taking paradigm itself.
11.5.2. Lynn and Vanhanen' s Findings
Lynn’s position is set forward in numer¬
ous papers and in two books, both coau¬
thored with the finnish economist Tutu
Vanhanen.189 The second book is basically an
expansion of the first, with added data and
analyses, none of which changes any of the
conclusions of the first. I shall concentrate
on the presentation in the second book, IQ
and Global Inequality. Specific papers will
be cited only if needed to make a point.
Lynn and Vanhanen regard intelligence as
being largely a genetically determined trait.
Although they do not say so precisely, their
argument makes no sense unless they also
regard the genetically defined reaction range
as being narrow, as they are not optimistic
about the effects of such things as improved
education.
They believe that intelligence makes a
contribution to indicators of national health
and wealth both through its interaction with
education and directly. Their argument is
that intelligent people are more educable
(which is certainly true) and that, in addi¬
tion to the benefits of education, intelligent
people are generally better reasoners, and
hence more able to deal with the problems
of a complex society. Lynn, in his numer¬
ous publications, has also made it clear
189 Lynn & Vanhanen, 2002, 2006.
438
HUMAN INTELLIGENCE
Table 11.5. Correlations between IQ estimates and
selected indices of national well-being
Correlation with
Index of Quality of Life
IQ Estimate
Gross national income/person (adjusted for
local purchasing power)
.684
Adult literacy rate
.642
Tertiary education
.746
Life expectancy
■m
Democratization Index
.568
Note: Data selected from Lynn and Vanhanen, 2006, Table 6.1.
Figures shown indicate the data based on 113 observed points.
that he regards intelligence test scores as
equally valid across nations, including those
nations usually referred to as “developing”
or "Third World.” For instance, he and Van¬
hanen include in their database test scores
from the United States (mean IQ = 98), the
United Kingdom (100), the Marshall Islands
(84), China (105), Ghana (71), and Equatorial
Guinea (59, to be discussed later).190 Their
statistical treatment gives all points equal
validity.
Lynn and Vanhanen obtained estimates
of IQs for each nation by searching vari¬
ous publication sources. This produced data
for 113 countries. They estimated the IQs
for another seventy-nine countries by using
"comparison IQs” of "neighboring or other
comparable countries.” This provided them
with three sets of data, 113 direct observa¬
tions, 79 imputed data points, and a total of
192 data points, obtained by combining the
first two sets.
In their main analyses Lynn and Van¬
hanen considered the relation between the
IQ measures and five measures of, as they
put it, “[t]he quality of human condi¬
tions.” The measures of quality they used
were gross national income per capita
(corrected for local purchasing power),
the adult literacy rate, the fraction of
the population enrolling in tertiary educa¬
tion (college/university and beyond), life
expectancy, and an index of democratiza¬
tion developed by Vanhanen from previous
work. This index can be thought of as a
measure of the extent to which citizens of
a country are free to participate in polit¬
ical life. The highest marks on the index
(forty and above) were assigned to Belgium,
Denmark, the Netherlands, and Switzer¬
land. Most large European and North Amer¬
ican countries scored in the mid-thirties.
Dictatorships that (as of 2010) enforced a
one-party state, such as China, Cuba, Kaza¬
khstan, and North Korea, scored zero.
The variables were merged into a single
quality of human conditions (QHC) index.
Lynn and Vanhanen then calculated regres¬
sion equations predicting the QHC index or
one of its components from the IQ scores, in
each of their three data sets. Additional anal¬
yses were also carried out, relating IQ scores
to other social indices, such as measures of
undernourishment and economic freedom.
None of these analyses made any difference
to Lynn and Vanhanen’s conclusions, so I
will concentrate on their central points.
Table 11.5 shows the correlations between
the IQ estimates and the five indices of qual¬
ity of human conditions. The correlations
range from a high of .773 for life expectancy
to a low of .568 for democratization. Given
this level of correlation with the individual
variables, it is inevitable that there will be
a high correlation with the composite QHC
index, and there is, .805 in the 113 observed
data points.191 The data presented is typi¬
cal of a number of other analyses, all of
190 Lynn & Vanhanen, 2006, Table 4.3.
191 Lynn & Vanhanen, 2006, Table 7.1.
THE DEMOGRAPHY OF INTELLIGENCE
439
which make the same point. On a national
basis the IQ estimate is a good predictor of
socioeconomic indicators of quality of life.
Lynn and Vanhanen reach a stronger con¬
clusion, about causation, which I will discuss
later.
Lynn and Vanhanen present one other
set of analyses that are relevant to our con¬
cerns. They group countries by geographic
area, note the predominant racial groups
in those areas, and then calculate a repre¬
sentative value for IQ by race, based on
this data. For instance, sub-Saharan Africans
(i.e., Black, “Negroid” in some of Lynn’s
writings) are assigned a typical IQ of 67,
East Asians receive 105, 99 for Europeans,
and 90 for Southeast Asians. They further
show that by assigning intermediate values
to “mixed-race” groups, such as the Mestizos
in some Latin American countries and “col¬
oreds” in South Africa, they can produce an
estimated IQ for the Latin American nations
that predicts observed values.
Lynn and Vanhanen, and Lynn in other
writings, draw three conclusions from their
research.
1. The substantial correlations between IQ
scores and various socioeconomic indi¬
cators shows that differences between
countries in wealth, health, and to some
extent happiness are very largely caused
by differences in intelligence.
2. The differences in intelligence between
nations are largely due to differences in
the racial composition of the national
and regional population.
3. While the differences have some envi¬
ronmental roots, the major differences
are due to genetic inheritance. Nutri¬
tional differences are the most impor¬
tant environmental factors. They do
allow some influence for education, but
believe that it is not likely to be large.
Conclusions like this are bound to be con¬
troversial. Therefore, it is worth taking a
closer look at the Lynn- Vanhanen data, and
then looking at another analysis aimed at the
same questions.
The Lynn and Vanhanen database for
intelligence is highly suspect in one way, and
accurate enough in another.
In both their publications Lynn and Van¬
hanen disregarded any question about the
validity of various intelligence tests across
different countries and cultures, and ignored
considerations of differential sampling. In
some cases their criteria for inclusion can
only be described as a blunder. For instance,
their data point for Equatorial Guinea, said
to have a national IQ of 59 (which would
be well into the mentally retarded range
in a developed country), was taken from a
report192 that, while it did discuss Africa,
stated clearly that the IQ 59 figure referred
to children in a school for the developmen-
tally disabled in Madrid! Many of the sam¬
ples were grossly unrepresentative, and the
criteria for selection of studies to include in
their work were not made clear.
These concerns have led critics to claim
that the data for developing countries is
both unrepresentative and extremely sus¬
pect. In order to test this hypothesis Werner
Wittmann and I conducted separate anal¬
yses of the data points for developed and
developing countries, as presented in Lynn
and Vanhanen’s 2002 book.193 We found that
the residuals (i.e., the extent to which their
predictions missed their targets) were much
greater for developing than for developed
countries. In much more detailed reviews
which found many more problems of selec¬
tivity and scholarship in the Lynn and Van¬
hanen data set, Wicherts, Dolan, and van
der Maas raised their estimate of the median
sub-Saharan African’s IQ of Lynn’s estimate
of 67 (an “educable mental retardate” in the
jargon of clinicians in the developed coun¬
tries) to 82. 194 Wicherts and his colleagues
192 Fernandez-Ballesteros et al., 1997.
193 Hunt & Wittmann, 2008.
194 Wicherts, Dolan, & van der Maas, 2010a. Wicherts,
Dolan, and van der Maas based their calculations
on a large number of studies using the Raven tests,
in order to avoid problems involved in comparing
different tests Lynn and Meisenberg (2010) claimed
that the studies Wicherts and his colleagues had
used were not representative, and offered their own
survey, which reinforced their original point. In a
second paper Wicherts, Dolan, and van der Maas
440
HUMAN INTELLIGENCE
pointed out that if you go backward in
time by correcting for the increases in IQ
scores over the past fifty years, this is the
modern equivalent of the IQ of the Dutch
just after World War II. The Wicherts
and colleagues work showed that Lynn and
Vanhanen’s research was highly biased
toward underestimation of countries in sub-
Saharan Africa. Given that this was true of
Africa, one has to be more than a little suspi¬
cious of the accuracy of their data for other
developing countries.
In spite of these concerns, Lynn and
Vanhanen's conclusions about the correla¬
tions between IQ estimates and measures of
social well-being are probably correct. The
spreads between nations in terms of various
measures of social well-being are so wide
that it is fairly easy to find a correlation
between them. Also, as measures of social
well-being are markedly collinear, a predic¬
tor that tracks one outcome measure will
probably track them all. These points were
illustrated by two researchers from George
Mason University, Deborah Whetzel and
Michael McDaniel. They assigned a value of
90 to all reported national IQs below 90 in
the 2002 database. This arbitrary assumption
was higher than the Wicherts and colleagues
estimate, which was not available at the time
that Whetzel and McDaniel did their anal¬
ysis. All major trends in the 2002 data still
held. I am virtually certain that the same
thing would happen in the 2006 database,
although I do not know of any such analysis.
There is a simple reason for the fact
that correlations were obtained in spite
of inaccurate data. Lynn and Vanhanen's
results were driven by the sharp dif¬
ferences in both intelligence test scores
and measures of social well-being between
the European-North American-Northeast
Asian (ENAMA] regions of the world and
“the rest,” largely sub-Saharan Africa and
South Asia (SASA). The South America
(2010b] showed that Lynn and Meisenberg’s tech¬
niques for selecting papers to cite were not related
to measures of representativeness in the original
research, and that they appeared to introduce a sub¬
stantial bias toward underestimating the cognitive
capacities of sub-Saharan populations.
and the Middle East regions (SAME] gen¬
erally fall somewhere between the other
two regions, in terms of both intelligence
test scores and indicators of well-being.
However, substantial correlations, although
lower than Lynn and Vanhanen reported, do
hold within the developed countries, where
the intelligence estimates are probably far
more accurate.195
11.5.3. A Much Closer Look:
Rindermanris Analyses
Lynn and Vanhanen rested their case largely
on bivariate correlations between variables,
which is a suspect procedure. They con¬
cluded that gross measures of educational
accomplishment, such as the literacy rate
or the number of people involved in ter¬
tiary education, were less important than
intelligence as predictors of national well¬
being. A more detailed analysis of their own
data, presented in Table 11.6, raises ques¬
tions about this conclusion. The entries in
Table 11.6 indicate the fraction of variance
in the target variable (GNI per capita, life
expectancy, or Democracy Index) that can
be associated with various possible causal
factors. For example, the first entry in the
first row, .47, indicates that 47% of the vari¬
ance in GNI per capita, across nations, can
be statistically predicted from variations in
the IQ estimate. That is impressive. The
second entry in the first row, .07, modifies
the impression considerably. It shows that
after allowing for associations between IQ
and GNI that can be predicted from rate
of participation in tertiary education, only
7% of the variance in GNI is associated with
IQ. The third entry in the row, .29, shows
that 29% of the variance in GNI is associated
with national differences in participation in
tertiary education, after having allowed for
differences in IQ.
Analyses such as those in Table 11.6 show
the difficulty of interpreting bivariate cor¬
relations in situations where the variables
involved are highly collinear. Suppose we
regard participation in tertiary education as
195 Hunt & Wittmann, 2005.
THE DEMOGRAPHY OF INTELLIGENCE
441
Table 11.6. Squared correlations and partial correlations between
target variables and IQ scores and education variables
Variable
IQ
IQlTertiary
TertiaryttQ
IQiLiteracy
Literacy\IQ
GNI/capita
O.47
0.07
0.28
0.25
0.07
Life expect.
0.60
0.29
0.07
0.35
0.19
Democracy
O.32
0.01
0.25
0.11
0.09
Note: The calculations are based on the data presented in Lynn & Vanhanen,
2006, Table 6.1. GNI - gross national income; Life expect. - life expectancy
at birth; Democracy - index of democratic institutions in a nation, developed
by Vanhanen; Tertiary - fraction of population with some post-secondary
education; Literacy - literacy rate.
an indicator of the general state of educa¬
tion in a country. One could argue that if
the population as a whole has generally high
intelligence the country will have a good
school system, and as a result a high level
of income per person. Or one could argue
that because a country has a good school
system the people will acquire high intel¬
ligence, and thus have high incomes. Or,
as is more likely, there may be reciprocal
influences between wealth, intelligence, and
education.
We now turn to a study that dealt with
such collinearities and ambiguities.
Heiner Rindermann, of the Otto-von-
Guericke University in Germany, has car¬
ried out multivariate investigations of
national cognitive skills, wealth, and well¬
being much further, and in a much more
sophisticated way, than did Lynn and Van¬
hanen (or than I have done in my reanalysis
of their data here.).196 Lynn and Vanhanen
paid relatively little attention to interna¬
tional assessments of educational data, such
as the PISA studies, except to use them
to validate their national intelligence esti¬
mates. (In general the correlations between
the educational assessment data and the
Lynn and Vanhanen intelligence estimates
are in the .8-9 range.) As Rindermann
notes, researchers in intelligence and edu¬
cation use different words, but they all wind
up generating tests that evaluate reasoning
196 Rindermann, 2007.
and knowledge, albeit with somewhat dif¬
ferent balances between the two.
The educational assessment data suffers
from the disadvantage that it is restricted to
children and adolescents, but it has several
counterbalancing advantages. Compared to
the intelligence data the educational sam¬
ples are much larger and more representa¬
tive of the countries involved. Where the
samples appear to be unrepresentative, as in
the case of differential participation rates by
students across countries, statistics on the
manner of unrepresentativeness are often
available. These statistics can be used to
make a revised estimate of a hypothetical
national average. For example, if a country
has low participation rates because it pro¬
vides schools for only some of its children,
the national estimate, based on the perfor¬
mance of schoolchildren, can be adjusted
downward. Within a study, the same tests
or tests that are "controllably different,” as
in the case of translations, are used in each
country.
There is another important advantage to
using educational rather than intelligence
test data. In several countries comparable
educational surveys have been carried out at
different points in time. Longitudinal anal¬
ysis can then be applied to determine what
variables at time 1 best predict the state of a
country at time 2. If the time period between
evaluations is close to a generation (thirty
years for humans), the best data for this
prediction are the cognitive skills of school
442
HUMAN INTELLIGENCE
children at time 1, for they are the ones who
will he running the society at time 2.
Rindermann first showed that there is
a huge general factor for 'national cogni¬
tive skill/’ calculated across countries. The
Lynn and Vanhanen estimates load heav¬
ily on this factor, although they are not the
best marker.197 The centrality of this factor
is amazing; the range of loadings is from .97
to 1.0I The reason that this is a bit surpris¬
ing is that some of these tests were intended
to cover special topics, such as reading or
mathematics. On a cross-national basis, if
the children read well they do math well
(and vice versa)/98
Rindermann then created a measure of
national cognitive skill, based on the above
analyses. He dropped the Lynn and Van¬
hanen data set from this analysis but, as
the high factor loadings show, it could not
have made much difference if it had been
included/99 On a regional basis Rindermann
reconstructed, with much better data, the
ENAMA-SAS A-SAME groupings that Lynn
and Vanhanen found. The lowest scores
are found in sub-Saharan Africa. The only
exception is Haiti in the Caribbean, where
scores are also quite low. The two interme¬
diate groups of scores are found in South
Asia, the Middle East and North Africa, and
in South America. The highest scores are
found in northeast Asia, Europe, and North
America.
Rindermann has reported a number of
correlations between his cognitive skills
index and indices of economic and politi¬
cal well-being. Some mirror the Lynn and
Vanhanen data. For instance, the corre¬
lation with gross domestic product per
197 Rindermann, 2007, Table 1.
198 This does not mean that there are no variations in
relative skill. Some countries are “tilted” toward
high math scores or high language scores. See
Wittmann, 2005. The same countries differ in level
of performance, as defined by a mixture of math¬
ematics, reading, and other scores. In a statistical
analysis variations in level are much greater than
variations in tilt.
199 The factor loading for the Lynn and Vanhanen esti¬
mates was .96. Rindermann’s reason for dropping
this data set was that it was a potpourri of mea¬
sures taken at different times, and he wished to do
longitudinal analyses.
capita (GDP/c) was .60, close to the Lynn-
Vanhanen estimate. Other indicators show
similar trends. The correlation with an index
of economic freedom was .52, and with eco¬
nomic growth .44.
The correlation between the cognitive
skills index and fertility (children per
woman) was strongly negative, —.73. This is
an important finding, because it may point
the way to future changes in cognitive skills.
As was pointed out in Chapter 9, there
is a negative correlation between indices
of intelligence and family size. Although
family sizes are much larger in the devel¬
oping than in the industrially developed
countries, they are dropping as the devel¬
oping nations become urbanized, probably
because of many other social changes related
to increasing socioeconomic opportunities
for women. This may presage an increase
in intelligence in the developing countries
in coming generations. Only time will tell.
For those who like food for thought,
the correlation between national cognitive
skill and homicide rate was —.23, and
the correlation between cognitive skill and
the rate of solved homicide cases was
.32. Smart populations produce moderately
smart detectives.200
Because educational data was available
over time, Rindermann was able to conduct
causal analyses, in which he determined
the extent to which a social or psycholog¬
ical variable measured at time 1 predicts
social and psychological variables at time
2. His analysis relied on a technique called
cross-lagged analysis, which made it possible
to determine standardized path coefficients.
These statistics can be interpreted as mea¬
suring the extent to which a raise of one
standard deviation unit in a predicting vari¬
able will cause a raise, in standard deviation
units, of a predicted variable. The analysis
produced two interesting results.
The path coefficient between cognitive
ability, measured in the 1960-70 period, and
GDP/c, measured in 1998, was .29, which
indicates that raising academic achievement
by one deviation unit will “pay off’ by
200 Rindermann, 2008a, Figure 2.
THE DEMOGRAPHY OF INTELLIGENCE
443
increasing GDP/c by almost 30%, no small
thing. Going the other way, the path coeffi¬
cient from GDP/c in 1980 to cognitive abil¬
ity in the 1990s was .21, indicating a feedback
process. Investing in education does pay off
in the future.201 This conclusion has been
strengthened by subsequent analyses, which
showed that by far the best predictors of
schoolchildren's cognitive competences are
countries’ general adult educational level,
and the extent to which a country provides
for early educational experiences, in pre¬
school and kindergarten.202
A second set of path analyses focused on
the relation between the national level of
cognitive skill and political development.205
Cognitive abilities measured in the 1960-70
period predicted levels of democratization,
the rule of law, and political freedom in the
1990s. An interesting point was that in these
analyses sheer level of education, as indexed
by data on attendance at different levels,
was generally as good a predictor as cog¬
nitive abilities, which could be considered
a measure of educational accomplishment.
Path coefficients were higher for the polit¬
ical than for the economic indicators. An
increase of one standard deviation unit in
cognitive abilities in the 1960s and 1970s was
associated with a .71 deviation unit increase
in the political freedom index in the 1990s. In
this case there were no indications of recip¬
rocal influence. Political freedom in the early
period did not predict cognitive ability in the
later period.
Dictators have a problem. In order to
increase their nation’s GDP/c they should
invest in education. But if they do, it can
come back to haunt them.
11.5.4. Contrasting Lynns Work
and Rindermann s
Because Lynn’s work has been quoted
widely by some authors, it is worth look¬
ing at the differences between the Lynn and
Rindermann approaches.
201 Ibid., Figure 5.
202 Rindermann & Ceci, 2009.
203 Rindermann, 2008b.
Rindermann’s work makes a compelling
case that investment in education increases
national cognitive skills, which in turn
increase wealth, which allows for more
investment in education. This leaves open
the question of why some nations have got¬
ten on the track toward success, and others
have not.
Lynn and Vanhanen have no hesita¬
tion. The ENAMA-SASA-SAME group¬
ing suggests to them an obvious correla¬
tion with race, especially since the lowest
part of the SASA group is in sub-Saharan
Africa.204 They observe parallels with the
within-country ordering of intelligence by
racial/ethnic group: Asians, Whites, various
mixed racial/ethnic groups, and then sub-
Saharan Africans. They assign a major role
to genetics, and also allow a role for nutri¬
tion, especially if children's malnutrition is
extreme.
Rindermann' s causal analysis is straight¬
forward and believable. It accounts for a
great deal of the variance in such variables
as GDP/c, for the residual unexplained vari¬
ance amounts to less than 20% of the total
variance. Therefore, while you can always
explain a finding by unmeasured variables,
there is not much left to explain. I hope that
future studies explore the effect of some
physical environmental variables, such as
nutrition, but this, presumably, will come.
Lynn's explanations cannot be ruled out,
but he goes far beyond the data. To see this,
let us consider some alternatives.
11.5.5. CLoiv the International Differences
Arose : Two uJust So " Stories
Lynn and Vanhanen and Rindermann agree
that cognitive skills vary across nations.
Why? Lynn and Vanhanen, and Lynn in
other writings,205 have offered a distal expla¬
nation, which purports to explain how
national differences arose in the genotype
for intelligence. Their argument is that as
people migrated northward away from early
centers of civilization they encountered
204 Lynn & Vanhanen, 2006, Chapter 12.
205 Especially Lynn, 2006.
444
HUMAN INTELLIGENCE
progressively harsher conditions due to cold
and wide seasonal variations in the availabil¬
ity of food. Therefore, the selective pres¬
sure for high intelligence was stronger in
the high northern latitudes. As a result the
people in northern Europe and northeast
Asia, and populations derived from them,
such as the present North American
and Australia-New Zealand populations,
became smarter than those who had
remained behind, in the more forgiving cli¬
mate of the subtropics.
This sort of explanation can be inter¬
preted as the regeneration of the justifica¬
tions for European colonization. But an idea
is not wrong just because the Victorians (and
Herodotusl) thought it to be true. There
have been attempts to fill out and mod¬
ify Lynn’s hypothesis. One pair of inves¬
tigators has reported correlations between
Lynn’s IQ estimates and two graduate stu¬
dents' estimates of the skin colors of peo¬
ple in various countries. (The students had
not been to the country in question.) Pos¬
itive findings were interpreted as support
for the hypothesis.206 Another investigator,
Sastsoi Kanazawa of the London School of
Economics, has argued that Lynn is right in
general, but that evolution was not due to
the cold or seasonal change per se. It was due
to the need to meet challenges that were not
present in the evolutionary homeland from
which mankind evolved, in the area occu¬
pied by present-day Ethiopia.207
Lynn supports his hypothesis by refer¬
ence to group differences in brain size, esti¬
mated from studies of cranial capacity. Even
if this data is accepted, the group differences
in brain size and the correlations between
brain size and intelligence are far too small
to account for the large racial/ethnic dif¬
ferences in IQ scores. Besides, the fossil
record indicates that human brain sizes have
decreased over the past 20,000 years.208 This
coincides with the expansion of Homo sapi¬
ens into the high latitudes following the
retreat of the glaciers. This is not in accord
2o6Templer & Arikawa, 2006.
207 Kanazawa, 2004, 2008.
208 Geary, 2005, p. 53.
with Lynn’s proposal that northern expo¬
sure, if you will, produced bigger brains.
Using the record of historical events,
rather than modern test scores, Jared
Diamond has proposed a quite different
scenario.209 He maintains that the problem¬
solving skills and general cultural knowl¬
edge of a people will be determined by the
extent to which they are forced to con¬
front other groups of people and exchange
ideas with them. Prior to relatively modern
times, geography posed substantial barriers
to such exchanges, for pre-industrial groups
found it difficult to move through unfamiliar
ecologies.
To take one of Diamond's favorite exam¬
ples, people who lived in the highlands of
the island of New Guinea developed agri¬
culture at about the same time it was devel¬
oped in Mesopotamia. The New Guinea
highlanders were isolated from the coast
by harsh mountain and jungle conditions;
Mesopotamia was at the hub of potential
travel routes into China and India, Europe,
and northern Africa. Except for agriculture,
the New Guinea highlanders were stuck in
the Stone Age until they were contacted
in the 1930s. The development of agricul¬
ture in Mesopotamia was a major step in
the history of civilization.
According to Diamond the European-
northeast Asian peoples prospered intellec¬
tually because the major axis of the Eurasian
land mass, East-West, facilitated commu¬
nication, while the major axes of Africa
and the Americas, North-South, presented
major barriers to communication. The rel¬
evant communication took place over his¬
toric, not evolutionary, time. Diamond can
be read as stressing social evolution of
societies, not genetic evolution of individ¬
uals, as a major force in the development of
intelligence.
We have two truly orthogonal explana¬
tions: the East-West one and the North-
South one. Is there any evidence that can
help us decide?
Personally, I am more impressed by Dia¬
mond's emphasis on the spread of ideas over
209 Diamond, 1997.
THE DEMOGRAPHY OF INTELLIGENCE
445
historical time than by Lynn's ideas about
presumed selective pressures on the geno¬
type over evolutionary time. However, both
Lynn’s and Diamond’s analyses are "just so”
stories. Our present-day distribution of cog¬
nitive skills is the result of the unique history
of humans on Earth. We can develop stories
about how this distribution came to be. We
can show that some are not in accord with
the facts, but given only one history, over
evolutionary or historic time, it is extremely
unlikely that we will settle on one of sev¬
eral alternatives. What is the right one? We
could answer this only by a careful analy¬
sis of the cognitive behavior of populations
long since extinct. And this data we cannot
have.
11.5.6. Summing Up the Evidence for the
Worldivide Distribution of Intelligence
I have spent some time discussing Lynn and
Vanhanen’s work for two reasons. One is
that their results have been widely publi¬
cized but, to my concern, too often accepted
and too rarely carefully critiqued. The sec¬
ond one is that these findings are important.
Lynn and Vanhanen’s critics have tended
to reject their ideas out of hand, either
because they see IQ scores as irrelevant out¬
side of the industrial/post-industrial coun¬
tries or because they are concerned about
the weakness of many of the data points.
Such criticisms miss the mark. IQ scores are
not irrelevant as indicators of intelligence,
in the conceptual sense, in the developing
world; they are partially relevant. As Rinder-
mann showed, IQ scores are so highly cor¬
related with indices of educational achieve¬
ment that, statistically, one can serve as a
proxy for the other. Both are indicators of
the extent to which a national population
possesses the cognitive skills required in the
post-industrial world. These indicators are
important because the world is increasingly
dominated by the developed countries, so
the future of developing countries is tied to
the extent that they can acquire the skills
needed in post-industrial society.
Lynn and Vanhanen are investigating
a question that deserves serious study.
Therefore, it is not surprising that they
close both of their books with policy rec¬
ommendations. They believe that genetic
constrictions are so great that, although
modest improvements may be made by
improved nutrition, not very much can
be done to improve human capital in
the poorer nations. They are especially
pessimistic about the sub-Saharan African
nations. They say
The persistence of differences in intelligence
between nations is inemtable, and so, too,
will be the consequences: the persistence of
national differences in wealth. Or, as St.
John put it 2000 years ago, lipooryou have
always with you. "
Lynn and Vanhanen, 2006, p. 293
Lynn and Vanhanen’s conclusion is a
good example of a mindset for defeat. We
have a problem; it’s hard to solve; let’s
regard it as unsolvable. Rindermann’s anal¬
ysis shows that they have badly underval¬
ued the effects of investment in education.
I, personally, agree with Diamond that the
important thing is to have an exchange of
ideas in a population - something that is
related to but more than formal education.
So I have a very different view.
Can we do something to improve the
intelligence of the poorer, often desperately
poorer, peoples of the world? I wrote this
chapter the week that Barack Obama was
inaugurated President of the United States.
My question was answered by the signature
slogan of his campaign:
Yes we can1.
Barack Obama, in many speeches dur¬
ing the 2008 US presidential campaign.
Whether we actually will do anything to
improve cognitive skills on an international
scale is a policy decision, and as such beyond
the scope of this book
II. 6. Closing Comments on Group
Differences in Intelligence
Intelligence does vary with demography.
As we age, our information-processing
mechanisms deteriorate. The speed of
446
HUMAN INTELLIGENCE
mental processing slows, and our ability to
control attention lessens. Until great age,
though, these effects do not begin to match
the beneficial effects of practice. Your fifty-
five-year-old surgeon may not be as quick
as a physician just out of medical school,
but the experienced surgeon is likely to rec¬
ognize a situation more rapidly than a neo¬
phyte, and to be much more efficient at deal¬
ing with it. Your fifty-five-year-old airplane
pilot may not have the same capacity to con¬
trol attention that he or she had thirty years
ago, but the experienced pilot knows better
what to attend to, and has a library of expe¬
riences that can be drawn on. Fluid intelli¬
gence, Gf, may be a dramatic thing to study
in the laboratory, but in everyday life there
is a great deal to be said for Gc. The quickest
way to solve a problem is to recognize that
you have seen it before and already know
the answer.
These trends are important. The popula¬
tion is aging. Sheer economics dictates that
we will have to utilize an older workforce.
When life expectancies approach ninety,
neither the individual nor the society can
afford retirement at sixty. Designers of
future workplaces must take account of the
changing nature of cognition during the sec¬
ond half of life.
Men and women do not think alike; a
statement that may be a revelation to psy¬
chologists but certainly will not surprise
novelists. There is little, if any, average dif¬
ference in general reasoning between men
and women. In defiance to the novelists’
intuitions, when it comes to general reason¬
ing men are more variable than women.
Just as aging is a matter of degree, so
is thinking like a man or a woman. Over
and over again in this chapter I have used
the word “tend.” That is important. Women
tend to rely more on verbal thinking, men
tend to be better at spatial-visual reasoning.
There is plenty of overlap. Nevertheless, I
think that we will continue to see that most
helicopter pilots are men.
Why these differences occur is still a topic
of research. The brain size hypothesis put
forward so strongly by Lynn and Rushton
appears to me to be a nonstarter. Differ¬
ences in brain structure and hormonal influ¬
ence are far more likely causal factors. Given
the influence of prenatal and neonatal hor¬
mone circulation upon brain development,
these causes are probably intertwined. And
then there are the social causes - differen¬
tial experience and different values for social
roles. These differences have their influence;
and they have their reasons for being, both
in the past and in today’s society. My own
belief is that individuals should be as free as
possible to choose their social roles, within
the limits of their own capacity, and that
other people should respect those choices.
At this point we are getting into the realm
of social belief, rather than a discussion of
the science of human intelligence, so I will
say no more.
Suppose it were possible to take those
cognitive skills that are required to “make
it” in modern, post-industrial society and
to assign every person in the world a num¬
ber, reflecting how many of those skills he
or she had. (If we had the ultimate IQ
test, we could do this!) We would find that
within countries some racial/ethnic groups
had a lower distribution of numbers than
other groups had. As in the case of male-
female differences, there would be substan¬
tial overlap, although probably not as much
as the overlap in the distribution of numbers
between men and women. What would we
find across countries, and more particularly,
within the developed countries?
The northeast Asian and White eth¬
nic groups would have almost equal num¬
bers. Which group had the highest aver¬
age would probably depend upon whether
we weighed verbal skills more heavily than
spatial-mathematical skills or vice versa.
What is the best weighting? It depends upon
whether you are looking for the next Ein¬
stein or the next Shakespeare.
African-derived groups, "Blacks” in a jar¬
gon that ignores the fact that there are
Melanesian and Australian-Papuan groups
that are every bit as dark as Africans but are
genetically distant from them, would almost
certainly tend (that word again) to have
lower numbers on a test of skills required
to make it in post-industrial society. Some
THE DEMOGRAPHY OF INTELLIGENCE
447
of these skills are undoubtedly relevant to
all societies, but we have little idea what
the mix is. Genetically, the African-derived
groups cluster differently from other groups,
but the classification is fuzzy, especially
when different racial/ethnic groups have
been living close to each other for gener¬
ations. Socially, the groups that have low
scores are generally those groups that have
the least contact with post-industrial soci¬
ety. This remark applies both to national
groups and to segments of a national soci¬
ety that are segregated from each other, or
from schooling and social opportunities, to
any degree.
Are these racial/ethnic distinctions
important? Yes, because to the extent that
members of a group wish to participate in
the industrial and post-industrial society,
they have to have the skills that society
utilizes. Are the distinctions inevitable?
Some professors and some politicians have
proclaimed, loudly, that they know the
answer to this question. However, those
people who are so certain seem to disagree
rather vehemently about whether the
answer is “yes” or “no.” I do not expect them
to agree with each other, any more than
I expect that the Pope and Shiite Islam’s
Grand Ayatollahs will agree on the nature
of God.
It could be that there are genetic con¬
straints that make inequality of cognition
across groups inevitable. This hypothesis
can never be ruled out, for doing so would
require proving the null hypothesis and, as
any good statistics instructor will tell you,
that is a logical impossibility. It is worth
remembering that no genes related to the dif¬
ference in cognitive skills across the various
racial and ethnic groups have ever been dis¬
covered. The argument for genetic differ¬
ences has been carried forward largely by
circumstantial evidence. Of course, tomor¬
row afternoon genetic mechanisms produc¬
ing racial and ethnic differences in intelli¬
gence might be discovered, but there have
been a lot of investigations, and tomorrow
has not come for quite some time now.
A number of social and environmental
causes for racial/ethnic differences in intelli¬
gence have been proposed. Some refer to the
physical environment: health, nutrition, and
exposure to atmospheric toxins. Some refer
to the social environment. These include
constraints imposed by the general society,
such as lack of adequate schools (particu¬
larly a problem in developing countries, but
certainly a problem in the industrially devel¬
oped ones) or discriminatory practices. In
other cases the constraints are imposed by
the group’s own mores. Examples would
be having more children than a family can
raise effectively, differential discrimination
within a group (think of fundamentalist
Islam’s view of women’s education), and
attitudes within a peer group toward short¬
term achievement as opposed to long-term
achievement in education.
The causes of differences in cognition
between old and young, men and women,
and various racial/ethnic groups should be
investigated. We have made legal and prac¬
tical distinctions between these categories
in the past, we do so now, and we proba¬
bly will do so in the future. Retirement reg¬
ulations, antidiscrimination policies, social
support for mothers and their children, and
different forms of affirmative action are all
part of a rational society. Demographic dif¬
ferences in intelligence are relevant to these
policies, regulations, and programs. It is best
if science informs policy makers, so inquiry
is appropriate. On the policy makers’ side,
scientists should not be restricted in their
inquiries because the results might be incon¬
venient. On the scientist’s side, the results
must be fully and honestly reported, regard¬
less of the scientists’ personal beliefs about
social policy.
CHAPTER 12
Summary and Prospectus
It’s tough to make predictions, especially
about the future.
- attributed to Yogi Berra,
American sports figure1
A colleague of mine once observed that the
ending sections of all scientific discussions
could be collapsed to a single word - much.
Much has been learned and much remains
to be learned. That is certainly true of intel¬
ligence. But let us be more specific.
12.1. A Summary
Scientific research on intelligence is just over
a century old. There has been progress; it
just has not been as rapid as we might have
hoped. A great deal of this research has
come in spurts. As in many other sciences,
these spurts have occurred when techno¬
logical or historic developments open up
new sources of relevant data. In research on
human intelligence that has happened three
i The quotation has also been attributed to the physi¬
cist and Nobel Prize winner Niels Bohr.
times. The development of group testing
initiated by the Army Alpha examination
in World War I showed that valid intelli¬
gence measures could be obtained without
expensive one-on-one interviews. The result
was an explosion of applications, and data,
in both academic and industrial settings. In
the 1930s the development of mathemati¬
cal techniques for handling multivariate data
made possible the evaluation of sophisti¬
cated psychometric models of human intelli¬
gence. The impetus of this development was
magnified with the development of comput¬
ers, which made computation-heavy meth¬
ods of analysis possible. Since about 1985
we have been busy digesting data from a
new source - the extensive neuroscientific
measurements that are now possible, These
include noninvasive techniques for looking
directly at indicators of functioning in the
normal brain. So what have we learned from
all this?
Chapter 1 presented a view of intelli¬
gence as a concept, and as it is assessed
by IQ and related tests. I argued, in agree¬
ment with several other authors (espe¬
cially Robert Sternberg, but with echoes of
448
SUMMARY AND PROSPECTUS
449
Raymond Cattell), that intelligence should
be thought of as the collection of cogni¬
tive skills and knowledge required to achieve
success in one’s society. Some of these skills
are common to all human societies, and
some of them have varying degrees of rele¬
vance depending on the society. Our present
cognitive tests evaluate some of the cogni¬
tive skills and knowledge needed in mod¬
ern industrial and post-industrial societies.
Without making any statement about the
moral values involved, the practical fact is
that these skills are important worldwide,
for there are far more societies interested in
becoming industrial or post-industrial cul¬
tures than there are developed societies
that want to revert to farming and herd¬
ing, let alone to being gleaners and hunter-
gatherers.
Chapter 2 introduced the tests. My exam¬
ples were certainly not exhaustive, but I
think that they will give readers an idea of
what the tests are like. In that discussion I
argued strongly, and not for the only time,
that the biggest restriction on intelligence
testing has been an acceptance of intelli¬
gence as something that has to be demon¬
strated within the “Drop in from the Sky"
test paradigm.
In Chapter 3 I presented an overview of
the distinction between behavioral theories
of intelligence and theories based on find¬
ings in the neurosciences. I concluded that
there is room for both of them. Chapter 4
expanded on this by discussing the concepts
of psychometric analysis and the models of
intelligence that have been developed using
psychometric tools. I made the (somewhat
controversial) assumption that the theory
you want depends upon what you want to
do with it. If the goal is to understand how
intelligence is derived from biological vari¬
ables, then the g-VPR model is useful. If the
goal is to understand how intelligence is used
in society, then the g and three-stratum Gc-
Gf models have much to recommend them.
The two goals are equally legitimate and can
be explored with equal scientific rigor.
Chapter 5 presented theories that are out¬
side the psychometric mode. I concluded
that, in spite of the publicity that it has
generated, there is little empirical support
for Howard Gardner's Multiple Intelligence
theory. I take no stand on how useful it is as
an inspiration for educators. Robert Stern¬
berg’s Theory of Successful Intelligence is
rooted in empirical data, and has clear con¬
nections to the use of information about
intelligence in society. Psychometric theo¬
ries are silent on this point. While many of
the individual studies in support of Stern¬
berg’s theory are, in my opinion, not as
strong as they have been made out to be,
the overall support cannot be disregarded. A
special strength of this theory is that it can
be expanded to evaluating behavior outside
of the testing situation.
In that chapter I also shed a small tear
for J. P. Das's PASS model of intelligence,
which may not have received the attention,
evaluation, and expansion that it should
have. The reasons seem to me more historic
and social than intellectual.
Chapter 6 discussed the relationship
between behavioral measures of human
information processing and intelligence, as
defined by test scores. It was concluded
that there are two major components of
the information-processing underpinnings
of intelligence: cognitive processing speed
and individual differences in the function¬
ing of the working memory-control of atten¬
tion complex. I questioned whether it is use¬
ful to try to break down these relationships
more finely. I think it is not; others may
disagree.
Chapter 7 discussed the brain structures
shown to be involved in intelligent action.
It is clear that there is no “center of intel¬
ligence” in the brain. Intelligent behavior
depends upon the functioning of a coordi¬
nated system involving many parts of the
brain. Certain regions are more involved in
cognitive behavior than others. The frontal
lobe and parts of the parietal lobe are espe¬
cially prominent in cognition. But the whole
brain is needed. And yes, bigger-brained
people are smarter, and smarter people have
brains that function more efficiently than
less-smart people. This is true in the literal
sense of efficiency; more intelligent people
show less brain metabolism, not more, than
450
HUMAN INTELLIGENCE
less intelligent people when both are pre¬
sented with the same problem.
Chapter 8 discussed the very strong evi¬
dence for a genetic contribution to intelli¬
gence, from Quantitative Behavior Genet¬
ics, and the lack of evidence from Molecular
Genetics that could identify the gene-
intelligence link. This is a point where time
will tell. We need more studies of the molec¬
ular genetic link. We do not need any more
studies trying to tie down the value of her-
itability coefficients in the general popula¬
tion. It was important to determine that the
heritability coefficient is not zero and that
it is not one. Where does it fall in between?
That number will vary over time and place.
Chapter 9 turned the argument around,
by presenting evidence for a myriad of
physical and social environmental variables
that affect intelligence. These include atmo¬
spheric lead, how much alcohol your mother
consumed when you were in utero, and how
good and how extensive your schooling was.
The point was made that intelligence devel¬
ops with cognitive engagement; some peo¬
ple acquire wisdom as they go through life,
others just have experiences.
Chapter 10 presented the benefits of intel¬
ligence, including its influence on educa¬
tional achievement, social achievement, and
such mundane things as income and occupa¬
tion. Contrary to widespread beliefs, intelli¬
gence does not just predict academic suc¬
cess. It predicts workplace and academic
success about equally well. The prediction
is not perfect, but it is better than any
other predictor we have today, most defi¬
nitely including personality tests. It was also
pointed out in this chapter that there is no
support whatsoever for the idea that the
very intelligent are social nerds, or that once
you have an adequate supply of intelligence
other personality traits become more impor¬
tant. Intelligent people, as indicated by test
scores, tend to be successful people, and
the more intelligence you have, the brighter
your prospects.
Chapter 11 discussed group differences
in intelligence. As we grow older we grow
slower, in a literal sense, and have less con¬
trol over the attention-working memory
complex. That is the bad news. The good
news is that older people know more. Elders
may solve by memory a problem that the
young have to work out for themselves. Rec¬
ollection and analogy to previously solved
problems are powerful aids in thinking.
There are differences in the ways that
men and women think, but they lie much
more along a verbal-spatial reasoning axis
than in general intelligence. For some
unknown reason, though, men are more
variable in their general reasoning powers
than women are.
With respect to racial/ethnic and national
distinctions: yes, there are considerable dif¬
ferences between some groups in the pos¬
session of general cognitive skills, the g
dimension of intelligence. These differences
are related to the socioeconomic and edu¬
cational achievements of group members,
on both the national and international lev¬
els. While some environmental causes for
group differences have been uncovered,
their effects are insufficient to account for
the observed differences between groups
and nations. Arguments have been made
assigning the causes to genetic differences,
but no example of an actual genetic mecha¬
nism causing group differences has ever been
found.
I closed Chapter 11 with a brief discus¬
sion of the problems involved in studying
this topic in the context of understandably
heated social concerns about the existence
of, or implications of, racial and ethnic dif¬
ferences in intelligence.
That is where we have been. Now where
are we going?
12.2. The Future
Yogi was right, predicting about the future
is difficult. I will try anyway.
Science advances when one of three
things happens. A scientific discovery may
open up new sources of relevant data.
Developments in solid-state, low-tempera¬
ture physics led to today’s imaging technol¬
ogy, which opened a huge new source of
data for the neuroscientists. Or a new idea,
SUMMARY AND PROSPECTUS
perhaps imported from another field, may
bring about a new way of looking at known
facts. Darwin’s Theory of Evolution, devel¬
oped after studying finches, turtles, and bee¬
tles, has profoundly altered the way we think
about differences in behavior between men
and women. Then there is the third way to
change. Sometimes a genius comes along,
someone who arrives at a time when the
old ideas are not quite working and who has
the creativity and breadth of thought to put
everything together. Newton, Darwin, and
Einstein qualify, but there aren't that many
other examples at their level of accomplish¬
ment.
There are two reasons for studying intel¬
ligence: to understand how individual dif¬
ferences in cognitive capability come about,
and to understand what those differences
mean in our society. The two purposes are
similar in two ways. They are both legiti¬
mate scientific goals, and they both require
some way of describing variations in intelli¬
gence.
I think that our present descriptions of
intelligence are pretty good. After all, they
are the culmination of over a century of
careful thought. It turns out that if you
want to look down from intelligence to the
brain, the g-VPR theory may be best; if you
want to think about how intelligence is used
in society, the g model, some version of
the three-stratum theory, or of Sternberg's
Theory of Successful Intelligence may be
most useful. This will depend upon the con¬
text in which the study is conducted. Note
the difference. Understanding how the brain
produces individual differences in cognition
is a single goal, and a single description may
be appropriate. Understanding how intelli¬
gence is used “in society” involves disparate
goals for disparate situations, and differ¬
ent descriptions may be needed. I do not
find this particularly bothersome. I do find
Procrustean attempts to make one model fit
all situations a bit annoying.
What about new sources of data? With
respect to the brain-mind connection we
have every reason to be optimistic. Advances
in imaging, biochemistry, and molecular
biology have and will continue to open up
451
huge new fields of data for psychological
inquiry. The biggest constraint on advances
here may be the expense involved. The
equipment and techniques are very expen¬
sive now, but that may change. One thing
will not change. Studies of individual differ¬
ences require lots of people, committing a
lot of their time. Time is money.
What about new data on the social devel¬
opment and uses of intelligence? I am a
bit optimistic here, even though spectacu¬
lar new machinery like that used by neu¬
roscientists is not on the horizon. There
are two things that have to be done. First,
we must break down the artificial boundary
between the study of personality and the
study of intelligence. In the world, the two
work together. There are examples where
this is already being done, notably in the
work of Ackerman and in the major longitu¬
dinal studies, such as the Seattle Longitudi¬
nal Study and the Study of Mathematically
Precocious Youth. I am optimistic that this
sort of research will develop. But then comes
the next issue.
We have to break out of the rigid
paradigm of thinking of intelligence as some¬
thing to be measured only in a formal test¬
ing session, the “Drop in from the Sky”
paradigm. That paradigm has its uses, and
it has had its day. We may be on the edge of
a technological breakthrough that will allow
us to do just this. In the modern world peo¬
ple leave a great many electronic tracks as
they move about. Banking and credit com¬
panies know a great deal about people's
movements and spending habits. Central¬
ized medical records are already widespread
in Europe, and the United States is mov¬
ing toward creating such databases. Studies
have already been reported in which peo¬
ple’s movement through space was tracked
by analyzing records of calls on cell phones.2
Special electronic badges have been used
to record nonverbal communication during
social mixers and business meetings.3 The
resulting data banks contain a tremendous
amount of information that, potentially,
2 Song et al., 2010.
3 Pentland, 2010.
HUMAN INTELLIGENCE
45^
could be used to construct models of how
intelligence is developed and used in the
world outside of both the laboratory and
the testing paradigm. The result might
be breakthroughs in our understanding of
the social aspects of intelligence that will
rival the breakthroughs achieved by the
use of imaging technologies and genet¬
ics to understand the biological aspects of
intelligence.
The use of central databases and tracking
technologies raise serious issues about pri¬
vacy and accountability. If society sees suf¬
ficient value in expanding our knowledge of
how cognition is used in everyday life, then
these problems can be solved. Will they be
solved? That depends on how well those
interested in the study of intelligence make
their case.
Then there is that last way of advancing -
finding the genius who puts it all together.
What we need is that person who will bring
his or her one-in-ten-million intelligence to
bear on the problem. According to the lat¬
est Digest of Educational Statistics , approx¬
imately 700,000 people graduate from US
four-year institutions annually. Let us dou¬
ble this to allow for geniuses outside the
United States. We can expect the necessary
genius to obtain a bachelor’s degree some¬
time in the next ten to fifteen years. Then
the trick will be to interest them in the
Psychology of Intelligence.
I hope this book will help.
References
Abad, F. J., Colom, R., Juan-Espinosa, M., & Gar-
da, L. F. (2003) Intelligence differentiation in
adult samples. Intelligence, 31 (2), 157-166.
Ackerman, P. L. (1986] Individual differences
in information processing: An investigation
of intellectual abilities and task performance
during practice. Intelligence , io, 101-139.
Ackerman, P. L. (1987) Individual differences in
skill learning: An integration of psychomet¬
ric and information processing perspectives.
Psychological Bulletin, 102 (1), 3-27.
Ackerman, P. L. (1996) A theory of adult intellec¬
tual development: Personality, interests, and
knowledge. Intelligence, 22 (2), 227-257.
Ackerman, P. L. (2000) Domain-specific knowl¬
edge as the “dark matter” of adult intelligence:
Gf/Gc, personality, and interest correlates.
Journal of Gerontology: Psychological Sciences,
55B, 69-84.
Ackerman, P. L., & Beier, M. E. (2001) Trait
complexes, cognitive investment, and domain
knowledge. In R. J. Sternberg & Elena L. Grig-
orenko (Eds.), The psychology of abilities, com¬
petences, and expertise. Cambridge: Cambridge
University Press, pp. 1-30.
Ackerman, P. L., & Beier, M. E. (2006) Determi¬
nants of domain knowledge and independent
study learning in an adult sample. Journal of
Educational Psychology, 98 (2), 366-381.
Ackerman, P. L., Beier, M. E., & Boyle, M. O.
(2002) Individual differences in working mem¬
ory within a nomological network of cogni¬
tive and perceptual speed abilities. Journal of
Experimental Psychology: General, 131 (3), 567-
589.
Ackerman, P. L., Beier, M. E., Boyle, M. O.
(2005) Working memory and intelligence: The
same or different constructs? Psychological
Bulletin, 131 (1), 30-60.
Ackerman, P. L., & Heggestad, E. D. (1997) Intel¬
ligence, personality, and interests: Evidence
for overlapping traits. Psychological Bulletin, 121
(2), 219-245.
Ackerman, P. L., & Rolfhus, E. L. (1999) The
locus of adult intelligence: Knowledge, abil¬
ities, and nonability traits. Psychology and
Aging, 14 (2), 314-330.
Alarcon, M., Plomin, R., Fulker, D. W., Corley,
R., & DeFries, J. C. (1999) Molarity, not mod¬
ularity: Multivariate genetic analysis of spe¬
cific cognitive abilities in parents and their
16-year-old children in the Colorado Adop¬
tion Project. Cognitive Development, 14 (1), 175—
Alfonso, V. C., Flanagan, D. C., & Radwan, S.
(2005) The impact of the Cattell-Horn-Carroll
theory on test development and interpretation
of cognitive and academic abilities. In D. P.
453
454
REFERENCES
Flanagan & P. L. Harrison (Eds.), Contempo¬
rary intellectual assessment: Theories, tests, and
issues. New York: Guilford, pp. 185-202.
Allahyar, M., & Hunt, E. (2003) The assessment
of spatial orientation using virtual reality tech¬
niques. International Journal of Testing, 3 (3),
263-275.
Amsel, E.; Langer, R., & Loutzenhiser, L. (1991)
Do lawyers reason differently from psychol¬
ogists? A comparative design for studying
expertise. In R. J. Sternberg & P. A. Frensch
(Eds.), Complex problem solving: Principles and
mechanisms. Hillsdale, NJ: Erlbaum, pp. 223-
250.
Anastasi, A. (1956) Intelligence and family size.
Psychological Bulletin, 53 (3), 187-209.
Anderson, D. R. (1998) Educational television
is not an oxymoron. Annals of the American
Academy of Political and Social Sciences, 557,
24-38.
Anderson, J. R. (1996) ACT: A simple theory of
complex cognition. American Psychologist, 51
(4) » 355“365-
Ando, J., Ono, Y., & Wright, M. J. (2001) Genetic
structure of spatial and verbal working mem¬
ory. Behavior Genetics, 31, 615-624.
Ankney, C. D. (1992) Sex differences in brain size:
The mismeasure of women too? Intelligence, 16
(3-4). 329“336-
Arija, V., Esparo, G., Fern andez-B allart, J., Mur¬
phy, M. M., Wachs, T. D., & McCabe, G.
(2001) Relation of maternal intelligence and
schooling to offspring nutritional intake. Inter¬
national Journal of Behavioral Development , 25
(5) . 444-449-
Ashton, M. C., & Lee, K. (2005) Problems with
the method of correlated vectors. Intelligence,
33 (4)/ 431_ 434-
Atkinson, R. C. (2005) College admissions and
the SAT: A personal perspective. Observer
(Magazine of the Association for Psychological
Science), 18 (5), 15-22.
Baare, W. E. C., Hulshoff Pol, H. E., Boomsma,
D. I., Posthuma, D., Geus, deE. J. C.,
Schnack, H. G., et al. (2001) Quantitative
genetic modeling of human brain morphology
variation. Cerebral Cortex, 11, 816-824.
Baddeley, A. D. (1968) A three-minute reason¬
ing test based on grammatical transformation.
Psychonomic Science, 10, 341-342.
Baddeley, A. D. (1986) Working Memory. Oxford:
Oxford University Press.
Baddeley, A. D., & Hitch, G. (1974) Working
memory. In G. Bower (Ed.), Recent advances
in learning and motivation (vol. 8) . New York:
Academic Press, pp. 47-90.
Baghurst, P. A., McMichael, A. J., Wigg, N. R.,
Vimpani, G. V., et al. (1992) Environmental
exposure to lead and children's intelligence at
the age of seven years. New England Journal of
Medicine, 327 (18), 1279-1284.
Balter, M. (2006) Bruce Lahn profile: Links
between brain genes, evolution, and cognition
challenged. Science, 314 (22 December), 1872.
Baltes, P. B., & Smith, J. (1990) Towards a psy¬
chology of wisdom and its ontogenesis. In
R. J. Sternberg (Ed.), Wisdom: Its origins,
nature, and development. Cambridge: Cam¬
bridge University Press, pp. 87-120.
Baltes, P. B., & Staudinger, U. M. (2000) Wisdom.
American Psychologist, 55 (1), 122-136.
Bamshad, M., Wooding, S., Salisbury, B. A., &
Stephens, J. C. (2004) Deconstructing the rela¬
tionship between genetics and race. Nature
Reviews: Genetics, 5 (August), 598-604.
Barnett, W. S. (1998) Long-term cognitive and
academic effects of early childhood education
of children in poverty. Preventive Medicine: An
International Journal Devoted to Practice and
Theory, 27 (2), 204-207.
Bar-On, R., Tranel, D., Denburg, N. L., &
Bechara, A. (2003) Exploring the neurological
substrate of emotional and social intelligence.
Brain: A Journal of Neurology, 126 (8), 1790-
1800.
Bartlett, J. (1980) Familiar Quotations (15th edi¬
tion, edited by E. Beck). Boston: Little Brown.
Batty, G. D., Der, G., Macintyre, S., & Deary,
I. J. (2006) Does IQ explain socioeconomic
inequalities in health? Evidence from a popu¬
lation based cohort study in the west of Scot¬
land. British Medical Journal, 332 (7541), 1-5.
Bayless, S., & Stevenson, J. (2007) Executive
functions in school-age children born very pre¬
maturely. Early Human Development , 84 (4),
247-254.
Bayley, N. (1968) Behavioral correlates of mental
growth: Birth to 36 years. American Psycholo¬
gist, 23 (1), 1-17.
Bayley, N. (2005) Bayley Scales of Infant and Tod¬
dler Development (3rd edition). San Antonio,
TX: Harcourt Assessment.
Beaujean, A. (2005) Heritability of cognitive abil¬
ities as measured by mental chronometric
tasks: A meta-analysis. Intelligence, 33 (2), 187—
201.
Becker, J. B., Berkley, K. J., Geary, N., Hampson,
E., Herman, J. P., & Young, E. (Eds.) (2008)
REFERENCES
455
Sex differences in the brain: From genes to behav¬
ior. Oxford: Oxford University Press.
Beer, J. S., Shimamura, A. P., & Knight, R. T.
(2004) Frontal lobe contributions to execu¬
tive control of cognitive and social behavior.
In M. S. Gazzaniga (Ed), The cognitive neuro¬
sciences (3rd ed.). Cambridge, MA: MIT Press,
pp. 1091-1104.
Benbow, C. P., Lubinski, D., Shea, D. L., &
Eftekhari-Sanjani, H. (2000) Sex differences
in mathematical reasoning: Ability at age 13.
Psychological Science, 11 (6), 474-480.
Benbow, C. P., & Stanley, J. C. (1996) Inequity
in equity: How “equity” can lead to inequity
for high-potential students. Psychology, Public
Policy, and Law, 2 (2), 249-292.
Berenbaum, S. A., & Resnick, S. (2007) The
seed of career choice: Prenatal sex hormone
effects. In S. I. Ceci & W. M. Williams (Eds.),
Why aren't there more women in science: Top
researchers debate the evidence. Washington,
DC: American Psychological Association, 147-
157-
Berg C., Hertzog, C., & Hunt, E. (1982) Age
differences in the speed of mental rotation.
Developmental Psychology, 18 (1), 95-107.
Berg, L. M. (2007) Socioeconomic factors and
substandard parenting. Social Service Review,
81 (3), 485-522.
Bethge, H.-J., Carlson, J. S., & Wiedl, K. H. (1982)
The effect of dynamic assessment procedures
on Raven Matrices performance, visual search
behavior, test anxiety, and test orientation.
Intelligence, 6 (1), 89-97.
Binet, A., & Simon, T. (1905) The develop¬
ment of intelligence in children. L’Annee Psy-
chologique, 11, 163-191.
Birnbaum, A. (1968) Some latent trait models and
their use in inferring an examinee's ability. In
F. M. Lord & M. R. Novick, Statistical theories
of mental test scores. Reading, MA: Addison-
Wesley, pp. 397-479.
Bittner, A. C., Carter, R. C., Kennedy, R. S., &
Harveson, M. M. (1986) Performance eval¬
uation tests for environmental research
(PETER): Evaluation of 114 measures. Percep¬
tual and Motor Skills, 63 (2, pt. 2), 683-708.
Bjerkedal, T., Kristensen, P., Skjeret, G. A., &
Brevik, J. L. (2007) Intelligence test scores
and birth order among young Norwegian men
(conscripts) analyzed within and between
families. Intelligence, 35 (5), 503-514.
Black, P., Harrison, G., Ley, C., Marshall, B.,
& Wiliam, D. (2003) Assessment for learning.
New York: Open University Press (McGraw-
Hill).
Black, P., & Wiliam, D. (1998) Assessment and
classroom learning. Assessment in Education, 5
(0, 7_17-
Blackwell, D. L., & Lichter, D. T. (2000) Mate
selection among married and cohabiting cou¬
ples. Journal of Family Issues, 21 (3), 275-302.
Blair, C. (2006) How similar are fluid cogni¬
tion and general intelligence? A developmen¬
tal neuroscience perspective on fluid cognition
as an aspect of human cognitive ability. Behav¬
ioral and Brain Sciences, 29 (2), 109-125.
Blair, C., Gramson, D., Thorne, S., & Baker,
D. (2005) Rising mean IQ: Cognitive demand
of mathematics education for young children,
population exposure to formal schooling, and
the neurobiology of the prefrontal cortex.
Intelligence, 33 (1), 93-106.
Bledsoe, C., Houle, R., & Sow, P. (2007) High
fertility Gambians in low fertility Spain:
The dynamics of child accumulation across
transnational space. Demographic Research, 16,
Article 12.
Blinkhorn, S. (2005) A gender bender. Nature, 438
(3 November), 31-32.
Boardman, D. (1990) Graphicacy revisited: Map¬
ping abilities and gender differences. Educa¬
tional Review, 42 (1), 57-64.
Bollen, K. A. (1994) Testing structural equation
models. Newbury Park, CA: Sage.
Boomsma, D., Busjahn, A., & Peltonen, L. (2002)
Classical twin studies and beyond. Nature
Reviews Genetics, 3, 872-882.
Boring, E. G. (1923) Intelligence as the tests test
it. New Republic, 35 (June 6), 35-37.
Born, M. Ph., & Lynn, R. (1994) Sex differences
on the Dutch WISC-R: A comparison with
the USA and Scotland. Educational Psychol¬
ogy, 14 (2), 249-254.
Bornkessel, I. D., Fiebach, C. J., & Friederici,
A. D. (2004) On the cost of syntactic ambi¬
guity in human language comprehension:
An individual differences approach. Cognitive
Brain Research, 21 (1), 11-21.
Bouchard, T. J., Jr. (1997) IQ similarities in twins
reared apart: Findings and responses to crit¬
ics. In R. J. Sternberg & E. Grigorenko (Eds.),
Intelligence , heredity, and environment. Cam¬
bridge: Cambridge University Press, pp. 126-
160.
Bouchard, T. J., Jr. (1999) Genes, environment,
and personality. In S. J. Ceci & W. M.
Williams (Eds.), The nature -nurture debate:
456
REFERENCES
The essential readings. Malden, MA: Blackwell,
pp. 97-103.
Bouchard, T. J., Jr. (2009). Strong inference: A
strategy for advancing psychological science.
In K. McCartney & R. Weinberg (Eds.), Expe¬
rience and development: A festschrift in honor
of Sandra Wood Scarr. London: Taylor and
Francis, pp. 39-59.
Bowen, W. G., & Bok, D. (1998) The shape of the
river: Long-term consequences of considering race
in college and university admissions. Princeton,
NJ: Princeton University Press.
Bowley, G. (2006) Out of his league. When Larry
Summers got to Harvard he saw lazy, leftish
professors inflating grades in what looked like
an outdated Yugoslav workers' co-operative.
The faculty saw a bumptious, ill-educated
boor hijacking their university. Graham Bow-
ley looks at why Harvard’s president had to
go. Financial Times , (13 May 2006), p. 19.
Bradley, R. H. (1993) Children's home envi¬
ronments, health, behavior, and intervention
efforts: A review using the HOME Inventory
as a marker measure. Genetic , Social, and Gen¬
eral Psychology Monographs, 119 (4), 437-490.
Bransford, J. D., Brown, A. L., & Cocking,
R. R. (Eds.) (1999) How people learn: Brain,
mind , experience and school. Washington, DC:
National Academy Press.
Brim, O. G., Jr. (1965) American attitudes toward
intelligence tests. American Psychologist, 20 (2)
(February 1965), 125-130.
Broadhurst, C. L., Cunnane, S. C., & Crawford,
M. A. (1998) Rift valley lake fish and shell¬
fish provided brain-specific nutrition for early
Homo. British Journal of Nutrition, 79 (1), 3-21.
Brody, L. C., & Blackburn, C. C. (1996) Nurturing
exceptional talent: SET as a legacy of SMPY.
In C. P. Benbow & D. Lubinski (Eds.), Intel¬
lectual talent: Psychometric and social issues.
Baltimore: Johns Hopkins University Press,
pp. 246-267.
Brody, N. (1992) Intelligence (2nd ed.) New York:
Academic Press.
Brody, N. (2003) Construct validation of the
Sternberg Triarchic Abilities Test: Comment
and reanalysis. Intelligence, 31 (4), 319-329.
Brody, N. (2004) What cognitive intelligence is
and what emotional intelligence is not. Psy¬
chological Inquiry, 15 (3), 234-238.
Brody, N. (2006) Geocentric theory: A valid alter¬
native to Gardner's theory of multiple intelli¬
gences. In J. A. Schaler (Ed.), Howard Gard¬
ner under fire: The rebel psychologist faces his
critics . Chicago: Open Court, pp. 73-94.
Brokaw, T. (1998) The greatest generation. New
York: Random House.
Bronfenbrenner, U., & Ceci, S. J. (1994) Nature-
nurture reconceptualized in developmental
perspective: A bioecological model. Psycholog¬
ical Review, 101 (4), 566-586.
Bronfenbrenner, U., McClelland, P., Wething-
ton, E., Moen, P., & Ceci, S. J. (1996) The
state of Americans. New York: Free Press.
Brothers, J. (1981) Large versus small families.
Good Housekeeping (February).
Brugman, G. M. (2006) Wisdom and aging. In
J. E. Birren & K. W. Schaie (Eds.), Handbook
of the psychology of aging (6th edition). Burling¬
ton, MA: Elsevier, pp. 445-476.
Brunner, M., Krauss, S., & Kunter, M. (2008)
Gender difference in mathematics: Does the
story need to be rewritten? Intelligence, 36 (5),
401-421.
Burden, M. J., Jacobson, S. W., Sokol, R. J., &
Jacobson, J. L. (2005) Effects of prenatal alco¬
hol exposure on attention and working mem¬
ory at 7.5 years of age. Alcoholism: Clinical and
Experimental Research, 29 (3), 443-452.
Burns, N. R., & Nettlebeck, T. J. (2003) Inspec¬
tion time in the structure of cognitive abilities:
Where does it fit? Intelligence, 31 (3), 237-255.
Burton, L., & Fogarty, G. J. (2003) The factor
structure of visual imagery and spatial ability.
Intelligence, 31 (3), 289-318.
Butterworth, B. (2006) Mathematical expertise.
In K. A. Ericsson, N. Charness, P. J. Feltovich,
& R. R. Hoffman (Eds.), The Cambridge hand¬
book of expertise and expert performance. Cam¬
bridge: Cambridge University Press, 553-568.
Cabeza, R., & Nyberg, L. (2000) Imaging cog¬
nition II: An empirical review of 275 PET
and fMRI studies. Journal of Cognitive Neu¬
roscience, 12 (1), 1-47.
Campbell, D. T. (1975) On the conflicts between
biological and social evolution and between
psychology and moral tradition. American Psy¬
chologist, 30 (12), 1103-1126.
Campbell, F. A., Pungello, E. P., Miller-Johnson,
S., Burchincal, M., & Ramey, C. T. (2001)
The development of cognitive and academic
abilities: Growth curves from an early child¬
hood educational experiment. Developmental
Psychology, 37 (2), 231-242.
Campbell, J. P., & Knapp, D. (Eds.), (2001)
Exploring the limits in personnel selection and
classification. Mahwah, NJ: Erlbaum.
Caplan, D., Waters, G., & Alpert, N. (2003)
Effects of age and speed of processing on
rCBF correlates of syntactic processing in
REFERENCES
457
sentence comprehension. Human Brain Map¬
ping, 19 (2], 112-231.
Caporael, L. R. (1976) Ergotism: The satan loosed
in Salem? Science, 192 (4234; 2 April), 21-26.
Capron, C., & Duyme, M. [1989) Assessment of
effects of socio-economic status on IQ in a
full cross-fostering study. Nature, 340 (623),
^ 552-554-
Capron, C., & Duyme, M. (1996) Effect of socio¬
economic status of biological and adoptive
parents on WISC-R subtest scores of their
French adopted children. Intelligence, 22 (3),
259-276.
Carlson, J. S., & Wiedl, K. H. (1992) Principles of
dynamic testing: The application of a specific
model. Learning and Individual Differences, 4,
153-166.
Carpenter, P. A., Just, M. A., & Shell, P. (1990)
What one intelligence test measures: A the¬
oretical account of processing in the Raven
Progressive Matrix Test. Psychological Review,
^ 97 (3). 404-43 !■
Carroll, J. B. (1993) Human cognitive abilities.
Cambridge: Cambridge University Press.
Carroll, J. B. (1995) Reflections on Stephen Jay
Gould’s The Mismeasure of Man [1981): A ret¬
rospective review. Intelligence, 21, 121-134.
Casey, M. B., Nuttall, R. I., Pezaris, E., &
Benbow, C. P. [1995) The influence of spatial
ability on gender differences in mathematics
college entrance test scores across diverse sam¬
ples. Developmental Psychology, 31, 697-705.
Cattell, R. B. [1940) Effects of human fertility
trends upon the distribution of intelligence
and culture. In Intelligence in nature and nur¬
ture: Thirty ninth yearbook of the national soci¬
ety for the study of education, Part I. Blooming¬
ton, IL: Society for the Study of Education,
pp. 221-233.
Cattell, R. B. (1957) Personality and motiva¬
tion: Structure and measurement. Yonkers- on -
Hudson, NY: World Book Co.
Cattell, R. B. (1971) Abilities: Their structure,
growth, and action. Boston: Houghton Mifflin.
Cattell, R. B. (1987) Intelligence: Its structure,
growth, and action. New York: North-Holland.
Cavallo, A., el-Abbadi, H., & Heeb, R. (1998) The
hidden gender restriction: The need for proper
controls when testing for racial discrimination.
In B. Devlin, S. E. Fienberg, D. P. Resnick, &
K. Roeder [Eds.), Intelligence, genes, and suc¬
cess: Scientists respond to The Bell Curve. New
York: Springer- Verlag, pp. 193-214.
Cawley, J., Conneeley, K., Heckman, J., & Vyt-
lactil, E. [1997) Cognitive ability, wages, and
meritocracy. In B. Devlin, S. Fienberg, D.
Resnick, & K. Roeder [Eds.), Intelligence, genes,
and success: Scientists respond to The Bell
Curve. New York: Springer-Verlag, pp. 179-
192.
Ceci, S. J. [1990) On intelligence. . . more or less:
A bio-ecological treatment on intellectual devel¬
opment. Englewood Cliffs, NJ: Prentice-Hall.
Ceci, S. J., & Liker, J. [1986) A day at the races: A
study of IQ, expertise, and cognitive complex¬
ity. Journal of Experimental Psychology: Gen¬
eral, 115, 255-266.
Ceci, S. J., & Liker, J. [1987) A day at the races:
A study of IQ, expertise, and cognitive com¬
plexity. Journal of Experimental Psychology:
General, 116, 90.
Ceci, S. J., & Papierno, P. B. (2005) The rhetoric
and reality of gap closing: When the “have
nots” gain but the “haves” gain even more.
American Psychologist, 60, 149-160.
Ceci, S. J., & Williams, W. M. (1997) Schooling,
intelligence, and income. American Psycholo¬
gist, 52 (10), 1051-1058.
Ceci, S. J., & Williams, W. M. [Eds.) (2007)
Why aren't there more women in science: Top
researchers debate the evidence. Washington,
DC: American Psychological Association.
Cerella, J. [1985) Information processing rates in
the elderly. Psychological Bulletin, 98, 67-93.
Cerella, J. (1990) Ageing and information -
processing rate. In J. Birren &K. Schaie [Eds.),
Handbook of the psychology of aging [third edi¬
tion). New York: Academic Press, pp. 201-222.
Chabris, C. F. (1999) Prelude or requiem for the
‘Mozart Effect’. Nature, 400 [26 August), 826-
827.
Chamorro-Premuzic, T., Furnham, A., & Ack¬
erman, P. L. (2006) Incremental validity of
the typical intellectual engagement scale as a
predictor of different academic performance
measures. Journal of Personality Assessment, 87
[3), 261-268.
Chen, J.-Q., & Gardner, H. (2005) Assessment
based on multiple-intelligences theory. In D.
P. Flanagan & P. L. Harrison [Eds.), Contempo¬
rary intellectual assessment: Theories, tests, and
issues. New York: Guilford, pp. 77-102.
Chiodo, L. M., Covington, C., Sokol, R. J., Han-
nigan, J. H., Jannise, J., Ager, J., Greenwald,
M., & Delaney-Black, V. (2007) Blood lead
levels and specific attention effects in young
children. Neurotoxicology and Teratology, 29
(5)- 538-546-
Choi, J., & Silverman, I. (2003) Processes underly¬
ing sex differences in route-learning strategies
458
REFERENCES
in children and adolescents. Personality and
Indimdual Differences, 34 (7), 1153-1166.
Chorney, M. J., Chorney, K., Seese, N., Owen,
M. J., Daniels, J., McGuffin, P., Thompson,
L. A., Detterman, D. K., Benbow, C., Lubin-
ski, D., Eley, T., & Plomin, R. (1998) A quan¬
titative trait locus associated with cognitive
ability in children. Psychological Science, 19 (3),
159-166.
Church, A. T., & Katigbak, M. S. (1991) Home
environment, nutritional status, and mater¬
nal intelligence as determinants of intellectual
development in rural Philippine preschool
children. Intelligence, 15 (1), 49-78.
Clark, M. J., & Grandy, J. (1984) Sex differ¬
ences in the academic performance of scholas¬
tic aptitude test takers. College Board Report
84-8. New York: College Entrance Examina¬
tion Board.
Clarke, J., & Haughton, H. (1975) A study of
intellectual impairment and recovery rates in
heavy drinkers in Ireland. British Journal of
Psychiatry, 126, 178-184.
Cohen, J. (1988] Statistical power analysis for the
behavioral sciences (2nd edition). Hillsdale, NJ:
Erlbaum.
Cole, M. (2005) Cross-cultural and historical per¬
spectives on the developmental consequences
of education. Human Development, 48, 195—
216.
Cole, M., Gay, J., Glick, J., & Sharp, D. W. (1971)
The cultural context of learning and thinking: An
exploration of experimental anthropology. New
York: Basic Books.
Collins, W. A., Maccoby, E. E., Steinberg, L.,
Hetherington, E. M., & Bornstein, M. H.
(2000) Contemporary research on parenting:
The case for nature and nurture. American
Psychologist, 55 (2), 218-232.
Colom, R., Abad, F. J., Garcia, L. F., & Juan-
Espinosa, M. (2002) Education, Wechsler's
Full Scale IQ, and g. Intelligence, 30 (5), 449-
462.
Colom R., Abad F. J., Quiroga, M-A., Shih, P. C.,
& Flores-Mendoza, C. (2008) Working mem¬
ory and intelligence are highly related con¬
structs, but why? Intelligence, 36 (6), 584-606.
Colom, R., Abad, F. J., Rebollo, I., & Shih, P. C.
(2005) Memory span and general intelligence:
A latent variable approach. Intelligence, 33 (6),
623-642.
Colom, R., Aluja-Fabregat, A., & Garcia-Lopez,
O. (2002) Tendencias de emparejamiento
selective en inteligencia, dureza de character,
extraversion e inestabilidad emocional. Psy-
chothemia, 14 (1), 154-158.
Colom, R., Escorial, S., & Rebollo, I. (2004) Sex
differences on the progressive matrices are
influenced by sex differences on spatial abil¬
ity. Personality and Individual Differences, 37
(6), 1289-1293.
Colom, R., Haier, R. J., Head, K., Alvarez- Linera,
J., Quiroga, M-A., Shih, P. C., & Jung, R. E.
(2009) Gray matter correlates of fluid, crys¬
tallized, and spatial intelligence: Testing the
P-Fit model. Intelligence, 37 (2), 124-135.
Colom, R., Juan-Espinosa, M., Abad, F., & Gar¬
cia, L. F. (2000) Negligible sex differences
in general intelligence. Intelligence, 28 (1), 57-
68.
Colom, R.; Jung, R. E., & Haier, R. J. (2006) Find¬
ing the g-factor in brain structure using the
method of correlated vectors. Intelligence, 34
(6), 561-570.
Colom, R., & Lynn, R. (2004) Testing the devel¬
opmental theory of sex differences in intelli¬
gence on 12-18 year olds. Personality and Indi¬
vidual Differences, 36 (1), 75-82.
Colom, R., Rebollo, I., Palacios, A., Juan-
Espinosa, M., & Kyllonen, P. C. (2004) Work¬
ing memory is (almost) perfectly predicted by
g . Intelligence, 32 (3), 277-296.
Committee on Prospering in the Global Econ¬
omy: An Agenda for American Science and
Technology. (2007) Rising above the gathering
storm: Energizing and employing America for
a brighter economic future. Washington, DC:
National Academies Press.
Connor, P. D., Sampson, P. D., Bookstein, F. L,
Barr, H. M., & Streissguth, A. P. (2001) Direct
and indirect effects of prenatal alcohol damage
on executive function. Developmental Neu-
ropsychology, 18 (3), 331-354.
Conway, A. R. A. (2005) Cognitive mechanisms
underlying intelligence: A defense of a reduc¬
tionist approach. In O. Wilhelm & R. W.
Engle (Eds.), Handbook of understanding and
measuring intelligence. Thousand Oaks, CA:
Sage, pp. 47-60.
Cook, P. J. & Ludwig, J. (1998) The burden
of acting black: Do black adolescents dispar¬
age academic achievement? In C. Jencks &
P. Meredith (Eds.), The black-white test score
gap. Washington, DC: Brookings, pp. 375—
400.
Cooke, A., Grossman, M., DeVita, C.; Gonzalez-
Atavales, J., Moore, P., Chen, W., Gee, J., &
Detre, J. (2006) Large scale network for sen¬
tence processing. Brain and Language, 96 (1),
H-36*
Coon, H., Fulker, D. W., DeFries, J. C., &
Plomin, R. (1990) Home environment and
REFERENCES
459
cognitive ability of 7-year-old children in the
Colorado Adoption Project: Genetic and envi¬
ronmental etiologies. Developmental Psychol¬
ogy, 26 (3), 459-468.
Corwyn, R. F., & Bradley, R. H. (2008) The
panethnic Asian label and predictors of
eighth-grade student achievement. School Psy¬
chology Quarterly, 23 (1), 90-106.
Counter, S. A., Buchanan, L. H., & Ortega, F.
(2005) Neurocognitive impairment in lead-
exposed children of Andean lead-glazing
workers. Journal of Occupational Environmen¬
tal Medicine, 47 (3), 306-312.
Court, H. (1983) “Sex differences in performance
on Raven’s Progressive Matrices”: A review.
Alberta Journal of Educational Research, 29 (1),
54-74-
Coutinho, M. J. & Oswald, D. P. (2005) State
variation in gender disproportionality in spe¬
cial education: Findings and recommenda¬
tions. Remedial and Special Education, 26 (1),
^ 7-15 ’
Cowley, J., Connelley, K., Heckman, J., & Vyt-
lacil, E. (1998] Cognitive ability, wages, and
meritocracy. In B. Devlin, S. E. Fienberg, D.
P. Resnick, & K. Roeder (Eds.], Intelligence,
genes , and success: Scientists respond to The
Bell Curve. New York: Springer-Verlag, 179—
192.
Coyle, T. R., & Pillow, D. R. (2008) SAT and
ACT predict college GPA after removing g.
Intelligence, 36 [6], 719-729.
Crede, M., & Kuncel, N. R. (2008] Study habits,
skills, and attitudes: The third pillar support¬
ing collegiate academic performance. Perspec¬
tives on Psychological Science, 3 (6), 425-453.
Cree, E. H. (Michael Levien, editor] (1982] Naval
Surgeon. New York: Dutton.
Cronbach, L. J. (1957] The two disciplines of sci¬
entific psychology. American Psychologist, 12,
671—684.
Cronbach, L. J. (1996] Acceleration among the
Terman males: Correlates in midlife and after.
In C. P. Benbow & D. J. Lubinski (Eds.), Intel¬
lectual talent: Psychometric and social issues.
Baltimore, MD: Johns Hopkins University
Press, 179-191.
Cronbach, L. J., & Snow, R. E. (1977) Apti¬
tudes and instructional methods: A hand¬
book for research on interactions. New York:
Irvington.
Cullen, M. J., Hardison, C. M., & Sackett,
P. R. (2004) Using SAT-grade and ability -job
performance relationships to test predictions
derived from stereotype threat theory. Journal
of Applied Psychology, 89 (2), 220-230.
Dai, X., Ryan, J. J., Paolo, A. M., & Harrington,
R. G. (1991) Sex differences on the Wechsler
Adult Intelligence Scale-Revised for China.
Psychological Assessment, 32 (2), 282-284.
Damasio, A. R. (1994) Descartes' error: Emo¬
tion, reason, and the human brain. New York:
Grosset/Putman.
Damasio, A. R. (1998) Emotion in the perspec¬
tive of an integrated nervous systems. Brain
Research Reviews, 26, 83-86.
Damasio, H., Grabowski, T., Randall, F., Gal-
aburda, A. M., & Damasio, A. R. (1994) The
return of Phineas Gage: Clues about the brain
from the skull of a famous patient. Science, 264
(20 May), 1102-1104.
Daneman, M., & Carpenter, P. A. (1980) Indi¬
vidual differences in working memory and
reading. Journal of Verbal Learning and Ver¬
bal Behavior, 19 (4), 450-466.
Daneman, M., & Carpenter, P. A. (1983) Indi¬
vidual differences in integrating information
between and within sentences. Journal of
Experimental Psychology: Learning, Memory,
and Cognition, 9, 561-584.
Daneman, M., & Merikle, P. M. (1996) Work¬
ing memory and language comprehension:
A meta- analysis. Psychonomic Bulletin, and
Review, 3 (4), 422-483.
Das, J. P., Naglieri, J. A., & Kirby, J. R. (1994)
Assessment of cognitive processes: The PASS the¬
ory of intelligence. Boston: Allyn and Bacon.
Deary, I. J. (1994) Sensory discrimination and
intelligence: Postmortem or resurrection?
American Journal of Psychology, 107 (1), 95-
115.
Deary, I. J. (2000) Looking down on human intelli¬
gence: From psychometrics to the brain. Oxford:
Oxford University Press.
Deary, I. J., Batty, G., & Gale, R. (2008) Bright
children become enlightened adults. Intelli¬
gence, 19 (1), 1-6.
Deary, I. J., & Der, G. (2005) Reaction time,
age, and cognitive ability: Longitudinal find¬
ings from age 16 to 63 years in representative
population samples. Aging, Neuropsychology,
and Cognition , 12, 187-215.
Deary, I. J., Der, G., & Shenkin, S. D. (2005) Does
mother’s IQ explain the association between
birth weight and cognitive ability in child¬
hood? Intelligence, 33 (5), 445-454.
Deary, I. J., & Egan, V., Gibson, G. J., Austin,
E. J., Brand, C. R., & Kellaghan, T. (1996)
Intelligence and the differentiation hypothe¬
sis. Intelligence, 23 (2), 105-132.
Deary, I. J., Ferguson, K. J., Bastin, M. E., Barrow,
G. W. S., Reid, L. M., Seckl, J. R., Wardlaw,
460
REFERENCES
J. M., & MacLullich, A. M. J. [2007) Skull size
and intelligence, and King Robert Bruce's IQ.
Intelligence, 35 (6), 519-528.
Deary, I. J., Irwing, P.; Der, G., & Bates, T. [2007)
Brother-sister differences in the g factor in
intelligence: Analysis of full, opposite-sex sib¬
lings from the NLSY 1979. Intelligence, 35 (5),
45i-456.
Deary, I. J., Stough, C. (1996} Intelligence and
inspection time: Achievements, prospects,
and problems. American Psychologist, 51 (6),
599-608.
Deary, I. J., Strand, S., Smith, P., & Fernandes, C.
(2007) Intelligence and educational achieve¬
ment. Intelligence, 35 (1), 13-21.
Deary, I. J., Whalley, L. J., Lemmon, H., Craw¬
ford, J. R.; & Starr, J. M. (2000) The stabil¬
ity of individual differences in mental ability
from childhood to old age: Follow-up of the
1932 Scottish Mental Survey. Intelligence, 28 (1),
49-55*
deFrias, C. M., Lovden, M., Lindenberger, U., &
Nisson, L.-G. (2007) Revisiting the dediffer¬
entiation hypothesis with longitudinal multi¬
cohort data. Intelligence, 35, 381-392.
Demetriou, A. (2002) The development of men¬
tal processing: Efficiency, working memory
and thinking. Monographs of the Society for
Research in Child Development, 67 (1), vii-
x54*
Demetriou, A., Kui, Z. X., Spanoudis, G.,
Christou, C.; Kyriakides, L., & Platsidou, M.
(2005) The architecture, dynamics, and devel¬
opment of mental processing: Greek, Chinese,
or Universal? Intelligence, 33 (2], 109-141.
Der, G., & Deary, I. J. (2006] Age and sex
differences in reaction time in adulthood:
Results from the United Kingdom Health and
Lifestyle Survey. Psychology and Aging, 21 (1),
62-73.
Derksen, J., Kramer, I., & Katzko, M. [2002) Does
a self-report measure for emotional intelli¬
gence assess something different than general
intelligence? Personality and Individual Differ¬
ences, 32 (1), 37-48.
Detterman, D. K. (1991) Reply to Deary and
Pagliari: Is g intelligence or stupidity? Intel¬
ligence, 15 (2), 251-255.
Detterman, D. K., & Daniel, M. H. (1989) Cor¬
relations of mental tests with each other and
with cognitive variables are highest in low IQ
groups. Intelligence, 13, 349-360.
Detterman, D. K., & Spry, K. M. (1988) Is it smart
to play the horses? Comment on “A day at the
races: A study of IQ, expertise, and cognitive
complexity." Journal of Experimental Psychol¬
ogy: General, 117 (1), 91-95.
Detterman, D. K., & Thompson, L. A. (1997)
What is so special about special education?
American Psychologist, 52 (10), 1082-1090.
Deutsch, G. K., Dougherty, R. F., Bammer, R.,
Siok, W. T., Gabrieli, J. D., & Wandell, B.
(2005) Children's reading performance is cor¬
related with white matter structure measured
by diffusion tensor imaging. Cortex, 41, 354-
363.
Devlin, B., Fienberg, S. E., Resnick, D. P., &
Roeder, K. (Eds.) (1998) Intelligence, genes, and
success: Scientists respond to The Bell Curve.
New York: Springer-Verlag.
Diamond, J. (1997) Guns, germs and steel: The
fates of human societies . San Francisco: Norton.
Dick, D. M., Aliev, F., Kramer, J., Wang, J. C.,
Hinrichs, A., Bertelsen, S., Kuperman, S.,
Schukit, M., Nurnberger, J., Jr., Edenberg,
H. J., Porjesz, B., Begleiter, H., Hesselbrock,
V., Goate, A., & Bierut, L. (2007) Associa¬
tion of CHRM2 with IQ: Converging evidence
for a gene influencing intelligence. Behavior
Genetics, 37 (2), 265-272.
Dickens, W. T., & Flynn, J. R. (2001) Heritability
estimates vs. large environmental effects: The
IQ paradox resolved. Psychological Review, 108
(3). 346-369-
Dobzhanzky, T. (1937) Genetics and the origin of
species. New York: Columbia University Press.
Dodrill, C. B. (1981) An economical method for
the evaluation of general intelligence in adults.
Journal of Consulting and Clinical Psychology ,
49 (5). 668-673.
Dodrill, C. B., & Warner, M. H. (ig88) Further
studies of the Wonder lie Personnel Test as a
brief measure of intelligence. Journal of Con¬
sulting and Clinical Psychology, 56 (1), 145—
H7-
Dolan, C. V., Colom, R., Abad, F. J., Wicherts,
J. M., Hessen, D. J., & Van Der Sluis, S. (2006)
Multi-group covariance and mean structure
modeling of the relationship between the
WAIS-III common factors and sex and edu¬
cational attainment in Spain. Intelligence, 34,
193-210.
Dolan, C. V., & Ha maker, E. L. (2001) Investi¬
gating Black- White differences in psychomet¬
ric IQ: Multi-group confirmatory factor anal¬
yses of the WISC-R and the K-ABC and a
critique of the method of correlated vectors.
In F. Columbus (Ed.), Advances in psychol¬
ogy research, Volume 6. Hauppauge, NY: Nova
Science Publishers, 31-59.
REFERENCES
461
Dolan, C. V., Roorda, W., & Wicherts, J. M.
(2004) Two failures of Spearman's hypothesis:
The GATB in Holland and the JAT in South
Africa. Intelligence, 32 (2), 155-173.
Dombrowski, S. C., Noonan, K.; & Martin, R. P.
(2007) Low birth weight and cognitive out¬
comes: Evidence for a gradient relationship
in an urban, poor, African American birth
cohort. School Psychology Quarterly, 22 (1), 26-
43-
Donders, J., Zhu, J., & Tulsky, D. (2001] Factor
Index score patterns in the WAIS-III standard¬
ization sample. Assessment, 8 (2), 193-203.
Duckworth, A. L., & Seligman, M. E. P. (2005]
Self-discipline outdoes IQ in predicting aca¬
demic performance of adolescents. Psycholog¬
ical Science, 16 (12), 939-944.
Duncan, J., Burgess, P., & Emslie, H. (1995) Fluid
intelligence after frontal lobe lesions. Neuro-
psychologia, 33 (3), 261-268.
Duncan, J., Emslie, H., Williams, P., & Johnson,
R. (1996] Intelligence and the frontal lobe: The
organization of goal directed behavior. Cogni¬
tive Psychology, 30 (3), 257-303.
Duncan, J., Seitz, R. J., Kolodny, J., Bor, D.,
Herzog, H., Ahmed, A., Newell, F. N., &
Emslie, H. (2000) A neural basis for general
intelligence. Science, 289 (5478), 457-460.
Dweck, C. S. (2006) Mindset: The new psychology
of success. New York: Random House.
Dye, M. W. G., Green, C. S., & Bavelier, D.
[2009) Increasing speed of processing with
action video games. Current Directions in Psy¬
chological Science, 18 (6), 321-326.
Earles, J. A., & Ree, M. J., (1992) The predic¬
tive validity of the ASVAB for training. Educa¬
tional and Psychological Measurement, 52, 721-
725*
Edwards, A. L. (1959] Objective approaches to per¬
sonality assessment. Princeton, NJ: Princeton
University Press.
Edwards, A. W. F. (2003) Human genetic diver¬
sity: Lewontin's fallacy. BioEssays, 25, 798-
801.
Eichenbaum, H. (2004) An information pro¬
cessing framework for memory representa¬
tion by the hippocampus. In M. S. Gaz-
zaniga (Ed.), The cognitive neurosciences (3rd
edition). Cambridge, MA: MIT Press, pp. 679-
690.
Elliott, C. D. (1986) The factorial structure and
specificity of the British Ability Scales. British
Journal of Psychology, 77 (2), 175-185.
Embretson, S., & Reise, S. P. (2000) Item response
theory for psychologists. Mahaw, NJ: Erlbaum.
Engle, R. W., Tuholski, S. W., Laughlin, J. E.,
& Conway, A. R. A. (1999) Working memory,
short-term memory, and general fluid intelli¬
gence: A latent variable approach. Journal of
Experimental Psychology: General, 128 (3), 309-
331*
Ericsson, K. A. (Ed.) (1996) The road to excellence:
The acquisition of expert performance in the arts
and sciences, sports, and games. Mahwah, NJ :
Erlbaum.
Ericsson, K. A. (2003) The search for gen¬
eral abilities and basic capacities: Theoreti¬
cal implications from the modifiability and
complexity of mechanisms mediating expert
performance. In R. J. Sternberg & E. L.
Grigorenko (Eds.), Abilities, competence, and
expertise. Cambridge: Cambridge University
Press, pp. 93-125.
Ericsson, K. A., Charness, N., Feltovich, P. J., &
Hoffman, R. R. (2006) The Cambridge hand¬
book of expertise and expert performance. Cam¬
bridge: Cambridge University Press.
Ericsson, K. A., Krampe, R. Th., & Tesch-Romer,
C. (1993) The role of deliberate practice in the
acquisition of expert performance. Psycholog¬
ical Review, 100 (3), 363-406.
Esbjorn, B. H., Hansen, B. M., Greisen, G., &
Mortensen, E. L. (2006) Intellectual develop¬
ment in a Danish cohort of prematurely born
preschool children: Specific or general diffi¬
culties? Journal of Developmental, Behavioral
Pediatrics, 27 (6), 477-484.
Evans, P. D., Gilbert, S. L., Mekel-Bobrov, N.,
Vallender, E. J., Anderson, J. R., Vaez-Azizi,
L., Tishkoff, S. A., Hudson, R. R, & Lahn,
B. T. (2005) Microcephalin, a gene regulat¬
ing brain size, continues to evolve adaptively
in humans. Science, 309 (9 September), 1717-
1720.
Fagan, J. F., & Holland, C. R. (2002) Equal oppor¬
tunity and racial differences in IQ. Intelligence,
30 (4), 361-387.
Fagan, J. F., Holland, C. R., & Wheeler, K. (2007)
The prediction, from infancy, of adult IQ and
achievement. Intelligence, 35 (3), 225-231.
Feldman, R. D. (1982) Whatever happened to the
quiz kids ? Perils and profits of growing up gifted.
Chicago: Chicago Review Press.
Fenner, J. N. (2005) Cross-cultural estimation
of the human generation interval for use in
genetics-based population divergence studies.
American Journal of Physical Anthropology, 128
(2), 415-423.
Fernandez-Ballesteros, R., Juan-Espinosa, M.,
Colom, R., & Calero, M. D. (1997) Contextual
462
REFERENCES
and personal sources of individual differences
in intelligence. In J. S. Carlson, J. Kingma, &
W. Tomic (Eds.), Advances in cognition and
educational practice: Reflections on the concept
of intelligence. London: JAI Press, 221-274.
Ferriman, K., Lubinski, D., & Benbow, C. P.
(2009) Work preferences, life values, and per¬
sonal views of top math/science graduates
and the profoundly gifted: Developmental
changes and gender differences during emerg¬
ing adulthood and parenthood. Journal of Per¬
sonality and Social Psychology, 97 (3), 517-532.
Finkel, D., & Pedersen, N. L. (2000) Contribution
of age, genes, and environment to the relation¬
ship between perceptual speed and cognitive
ability. Psychology and Aging, 15 (1), 56-64.
Finkel, D., & Pedersen, N. L. (2004) Processing
speed and longitudinal trajectories of change
for cognitive abilities: The Swedish adop¬
tion/twin study of aging. Aging, Neuropsychol¬
ogy, and Cognition, 11 (2, 3), 325-345.
Fish, J. M. (2002a) Race and intelligence: Separating
science from myth . Mahwah, NJ: Erlbaum.
Fish, J. M. (2002b) The myth of race. In J. M. Fish
(Ed.), Race and intelligence: Separating science
and myth. Mahwah, NJ: Erlbaum, pp. 113-141.
Fisk, A. D., & Schneider, W. (1983) Category and
word search: Generalizing search principles
to complex processing. Journal of Experimen¬
tal Psychology: Learning, Memory, Cognition, 9,
177-195-
Flynn, J. R. (1984) The mean IQ of Americans:
Massive gains 1932 to 1978. Psychological Bul¬
letin, 95 (1), 29-51.
Flynn, J. R. (1987) Massive IQ gains in 14 nations:
What IQ tests really measure. Psychological
Bulletin, 10 (2), 171-192.
Flynn, J. R. (1991) Asian Americans: Achievement
beyond IQ. Mahwah, NJ: Erlbaum.
Flynn, J. R. (2007) What is intelligence? Cam¬
bridge: Cambridge University Press.
Ford, R. T. (2008) The race card: How bluffing
about bias makes race relations worse. New
York: Farrar, Strauss, Giroux.
Fox, S., & Spector, P. E. (2000) Relation of
emotional intelligence, practical intelligence,
general intelligence, and trait affectivity with
interview outcomes: It’s not all just 'G.' Jour¬
nal of Organizational Behavior, 21 (Special
Issue), 203-220.
Freebody, P., & Tirre, W. C. (1985) Achieve¬
ment outcomes of two reading programmes:
An instance of aptitude-treatment interaction.
British Journal of Educational Psychology , 55 (1),
53-60.
Frey, M. C., & Detterman, D. K. (2004) Scholas¬
tic assessment or g? The relationship between
the scholastic assessment test and general cog¬
nitive ability. Psychological Science, 15 (6), 373—
378.
Friederici, A. D., Opitz, B., & von Cramon, D.
Y. (2000) Segregating semantic and syntactic
aspects of processing in the human brain: An
fMRI investigation of different word types.
Cerebral Cortex, 10 (7), 698-705.
Furnham, A. (2001) Self-estimates of intelligence:
Culture and gender difference in self and other
estimates of both general (g) and multiple
intelligences. Personality and Individual Differ¬
ences, 31 (8), 1381-1405.
Gajdusek, D. C., & Zigas, V. (1957) Degenerative
disease of the nervous system in New Guinea:
The endemic occurrence of kuru in the native
population. New England Journal of Medicine,
^ 257 0°)> 974“~978 •
Galton, F. (1869) Hereditary genius: An inquiry
into its laws and consequences. London:
MacMillan.
Ganis, G., Thompson, W. L., Mast, F., & Kosslyn,
S. M. (2004) The brain’s mind's images: The
cognitive neuroscience of mental imagery. In
M. S. Gazzaniga (Ed.), The cognitive neuro¬
sciences [yd edition ). Cambridge, MA: MIT
Press, pp. 931-941.
Gardner, H. (1983) Frames of mind: The theory of
multiple intelligences. New York: Basic Books.
Gardner, H. (1993a) Multiple intelligences: The the¬
ory in practice. New York: Basic Books.
Gardner, H. (1993b) Creating minds: An anatomy
of creativity seen through the lives of Freud, Ein¬
stein, Picasso, Stravinsky, Eliot, Graham, and
Gandhi. New York: Basic Books.
Gardner, H. (1999) Who owns intelligence?
Atlantic Monthly, 383 (February 1999), 67-76.
Gardner, H. (2006a) Replies to my critics: Reply
to Nathan Brody. In J. A. Schaler (Ed.),
Howard Gardner under fire: The rebel psychol¬
ogist faces his critics . Chicago: Open Court,
pp. 297-300.
Gardner, H. (2006b) On failing to grasp the core
of MI theory: A reply to Visser et al. Intelli¬
gence, 34 (5), 503-505.
Gardner, H., Kornhaber, M. L., & Wake, W. K.
(1996) Intelligence: Multiple perspectives . New
York: Harcourt Brace.
Garlick, D. (2002) Understanding the nature of
the general factor of intelligence: The role of
individual differences in neural plasticity as an
explanatory mechanism. Psychological Review,
109, 116-136.
REFERENCES
463
Geary, D. C. (2005) The origin of mind: Evolu¬
tion of brain, cognition, and general intelligence.
Washington, DC: American Psychological
Association.
Geary, D. C. (2007a) Educating the evolving
mind: Conceptual foundations for an evolu¬
tionary educational psychology. In J. C. Carl¬
son & J. L. Levine (Eds.), Educating the evolv¬
ing mind. Charlotte, SC: Information Age
Press, pp. 177-203.
Geary, D. C. (2007b) An evolutionary perspec¬
tive on sex differences in science and mathe¬
matics. In S. J. Ceci & W. M. Williams (Eds.),
Why aren't there more women in science: Top
researchers debate the evidence. Washington,
DC: American Psychological Association, 173—
188.
Geiser, S., 81 Studley, R. (2002) UC and the SAT:
Predictive validity and differential impact of
the SAT I and SAT II at the University of
California. Educational Assesment, 8 (1), 1-26.
Gelade, G. A. (2009) IQ, cultural values, and the
technological achievements of nations. Intelli¬
gence, 36 (6), 711-718.
Gentner, D., & Gentner, D. R. (1983) Flowing
waters or teeming crowds: Mental models of
electricity. In D. Gentner & A. L. Stevens
(Eds.), Mental models . Hillsdale, NJ: Erlbaum,
pp. 99-129.
Gigerenzer, G. (2000) Adaptive thinking: Ratio¬
nality for the real world. Oxford: Oxford Uni¬
versity Press, 2000.
Gignac, G. E. (2006) The WAIS-III as a nested
factors model: A useful alternative to the more
conventional oblique and higher-order mod¬
els. Journal of Individual Differences, 27 (2), 73-
86.
Goff, M., & Ackerman, P. L. (1992) Personality-
intelligence relations: Assessment of typical
intellectual engagement. Journal of Educa¬
tional Psychology, 84, 537-552.
Goldberg, R. A., Schwartz, S., & Stewart, M.
(1977) Individual differences in cognitive pro¬
cesses. Journal of Educational Psychology, 69
Oh 9_14-
Goldman, S. M., Tanner, C. M., Oakes, D.,
Bhudhikanok, G. S., Gupta, A., & Langston,
J. W. (2006) Head injury and Parkinson’s dis¬
ease risk in twins. Annals of Neurology, 60 (1),
„ 65_72-
Goleman, D. (1995) Emotional intelligence. New
York: Bantam Books.
Goodwin, D. K. (2005) Team of rivals: The political
genius of Abraham Lincoln. New York: Simon
and Schuster.
Gopnik, A., Meltzoff, A. N., & Kuhl, P. K. (1999)
The scientist in the crib: Mind, brains, and how
children learn. New York: Morrow.
Gorey, K. M. (2001) Early childhood educa¬
tion: A meta-analytic affirmation of the short-
and long-term benefits of educational oppor¬
tunity. School Psychology Quarterly, 16 (1),
^ 9_3°*
Gosso, M. F., van Belzen, M., de Geus, E. J.
C. , Polderman, J. C., Heutink, P., Boomsma,
D. I., & Posthuma, D. (2006) Association
between the CHRM2 gene and intelligence in
a sample of 304 Dutch families. Genes, Brain,
Behavior, 5 (8), 577-584.
Gosso, M. F., de Geuss, E. J. C., van Belzen, M.
J., Polderman, T. C., Heutink, P., Boomsma,
D. I., & Poshuma, D. (2006) The SNAP-25
gene is associated with cognitive ability: Evi¬
dence from a family-based study in two inde¬
pendent Dutch cohorts. Molecular Psychiatry,
11 (9), 878-886.
Gottfredson, L. S. (1986) Occupational aptitude
patterns match: Development and implica¬
tions for a theory of job aptitude require¬
ments. Journal of Vocational Behavior, 29
(monograph), 254—291.
Gottfredson, L. S. (1997) Why g matters: The
complexity of everyday life. Intelligence, 24 (1),
^ 79-U2-
Gottfredson, L. S. (1998) Jensen, Jensenism, and
the sociology of intelligence. Intelligence, 26
(3), 291-299.
Gottfredson, L. S. (2002) g: Highly general and
highly practical. In R. J. Sternberg & E.
L. Grigorenko (Eds.), The general factor of
intelligence: How general is it? Mahwah, NJ:
Erlbaum, pp. 331-381.
Gottfredson, L. S. (2003a) Dissecting practical
intelligence: Its claims and evidence. Intelli¬
gence, 31 (4), 343-398.
Gottfredson, L. S. (2003b) On Sternberg’s “Reply
to Gottfredson.” Intelligence, 31 (4), 415-424.
Gottfredson, L. S. (2005) Suppressing intelligence
research: Hurting those we intend to help. In
R. H. Wright & N. Cummings (Eds.), Destruc¬
tive trends in mental health: The well-intentioned
path to harm. New York: Routledge, 155-
186.
Gottfredson, L. S. (2007a) Applying double
standards to “divisive” ideas: Commentary
on Hunt and Carlson (2007). Perspectives on
Psychological Science, 2 (2), 216-220.
Gottfredson, L. S. (2007b) Innovation, fatal acci¬
dents, and the evolution of general intelli¬
gence. In M. J. Roberts (Ed.), Integrating the
464
REFERENCES
mind: Domain general vs domain specific pro¬
cesses in higher cognition. New York: Psychol¬
ogy Press, 387-425.
Gotz M. J., Johnstone, E. C., & Ratcliffe, S. G.
(1999) Criminality and antisocial behaviour in
unselected men with sex chromosome abnor¬
malities. Psychological Medicine, 29 (4}, 953-
962.
Gould, S. J. (1981] The mismeasure of man. New
York: Basic Books.
Gould, S. J. (1996) The mismeasure of man (Rev.
Ed.). Boston: Norton.
Grandy, J. (1999) GRE trends and profiles: Statis¬
tics about GRE © test examinees by gender , age,
and ethnicity (second edition). GRE No. 96-07.
Princeton, NJ: Educational Testing Service.
Gray, J. R., Chabris, C. F., & Braver, T. S. (2003)
Neural mechanisms of general fluid intelli¬
gence. Nature Neuroscience, 6, 316-322.
Greenfield, P. M. (1998) The cultural evolution of
IQ. In U. Neisser (Ed.), The rising curve: Long¬
term gains in IQ and related measures. Wash¬
ington, DC: American Psychological Associa¬
tion, 81-123.
Grigorenko, E. L., Meier, E., Lipka, J., Muhatt,
G., Yanez, E., & Sternberg, R. J. (2004) Aca¬
demic and practical intelligence: A case study
of the Yup'ik in Alaska. Learning and Individ¬
ual Differences , 14 (4), 183—207.
Grissmer, D., Flanagan, A., & Williamson, S.
(1998) Why did the Black- White score gap nar¬
row in the 1970s and 1980s? In C. Jenks & M.
Philips (Eds.), The Black-White test score gap.
Washington, DC: Brookings Institute, 182-
226.
Grudnik, J. L., & Kranzler, J. (2001) Meta-analysis
of the relationship between intelligence and
inspection time. Intelligence, 29 (6), 523-535.
Guerrero, L. M. K. (2009) The impact of lead con¬
tamination in children of Callao - Peru. Lima:
Fundacion Cayetano Heredia.
Guiso, L., Monte, F., Sapienza, P., &Zingales, L.
(2008) Culture, gender, and math. Science, 320
(30 May), 1164-1165.
Gur, R. C., & Gur, R. E. (2007) Neural substrates
for sex differences in cognition. In S. J. Ceci
& W. M. Williams (Eds.), Why aren't there
more women in science: Top researchers debate
the evidence. Washington, DC: American Psy¬
chological Association, pp. 189-198.
Gustafsson, J.-E. (1984) A unifying model for the
structure of intellectual abilities. Intelligence , 8
(3), 179-203.
Habib, R. J., McIntosh, A. R., & Tulving, E.
(2000) Individual differences in the functional
neuroanatomy of verbal discrimination learn¬
ing revealed by positron emission tomogra¬
phy. Acta Psychologia, 105 (2), 141-157.
Haier, R. J. (2009) Neuro-intelligence, neuro¬
metries, and the next phase of brain imaging
studies. Intelligence, 27 (2), 121-123.
Haier, R. J., & Benbow, C. P. (1995) $ex dif-
ferences and lateralization in temporal lobe
glucose metabolism during mathematical rea¬
soning. Developmental Neuropsychology , 11,
405-414.
Haier, R. J., Colom, R., Schroeder, D. H., Con¬
don, C. A., Tang, C., Eaves, E., & Head, K.
(2009) Gray matter and intelligence factors: Is
there a neuro-g? Intelligence, 37 (2), 136-144.
Haier, R. J., Jung, R. E., Yeo, R. A., Head, K., &
Alkire, M. T. (2004) Structural brain variation
and general intelligence. Neuroimage, 23, 425-
433-
Haier, R. J., Jung, R. E., Yeo, R. A., Head, K., &
Alkire, M. T. (2005) The new neuro an atomy
of general intelligence: Sex matters. Neuroim-
age, 25, 320-327.
Haier, R. J., Siegel, B. V., Neuchterlein, K. H.,
Hazlett, E., Wu, J. C., Paek, J., Browning,
H. L., & Buchsbaum, M. S. (1988) Cortical
glucose metabolic rate correlates of abstract
reasoning and attention studied with positron
emission tomography. Intelligence, 12 (2), 199-
2!?.
Haier, R. J., Siegel, B. V., Tang, C., Abel,
L., et al. (1992) Intelligence and changes in
regional cerebral glucose metabolic rate fol¬
lowing learning. Intelligence , 16 (3-4), 415-426.
Haier, R. J., White, N. S., Alkire, M. T. (2003)
Individual differences in general intelligence
correlate with brain function during nonrea¬
soning tasks. Intelligence, 31 (5), 429-441.
Hakstian, A. R., & Cattell, R. B. (1978) Higher-
stratum ability structures on a basis of twenty
primary abilities. Journal of Educational Psy¬
chology, 70 (5), 667-689.
Halpern, D. F. (1986) Sex differences in cognitive
abilities (1st edition). Hillsdale, NJ: Erlbaum.
Halpern, D. F. (2000) Sex differences in cognitive
abilities (2nd edition). Hillsdale, NJ: Erlbaum.
Halpern, D. F., Benbow, C. P., Geary, D. C.,
Gur, R. C., Hyde, J. S., & Gernsbacher, M. A.
(2007) The science of sex differences in science
and mathematics. Psychological Science in the
Public Interest, 6 (1), 1-51.
Halpern, D. F., & Tan, U. (2001) Stereotypes
and steroids: Using a psychobiosocial model to
understand sex differences in cognitive abili¬
ties. Brain and Cognition, 45 (3), 392-414.
REFERENCES
465
Harden; K. P., Turkheimer, E., & Loehlin, J. C.
(2007) Genotype by environment interaction
in adolescents' cognitive aptitude. Behawor
Genetics, 37, 273-283.
Harrs, M., & Hatano, G. [Eds.] (4999) Learning
to read and write : A cross-linguistic perspective.
New York: Cambridge University Press.
Hartigan, J. A., & Wigdor, A. K. (Eds.) (1989)
Fairness in employment testing: Validity gener¬
alization, minority issues, and the General Apti¬
tude Test Battery. Washington, DC: National
Academies Press.
Hartley, A. (2006) Changing role of the speed
of processing construct in the psychology of
aging. In J. E. Birren & K. W. Schaie (Eds.),
Handbook of the psychology of aging (6th edi¬
tion). Burlington, MA: Elsevier, pp. 183-207.
Hartmann, P., Kruuse, N. H. S., & Nyborg, H.
(2007) Testing the cross-racial generality of
Spearman's hypothesis in two samples. Intelli¬
gence, 35 (i), 47-57.
Hattori, K., & Lynn, R. (1997) Male-female dif¬
ferences on the Japanese WAIS-R. Personality
and Individual Differences, 23 (3), 531—533.
Haussmann, M., Slabbekoorn, D., Van Goozen,
S. H. M., Cohen-Kettenis, P. T., &
Guntiirkun, O. (2000) Sex hormones affect
spatial abilities during the menstrual cycle.
Behavioral Neuroscience, 114 (6), 1245-1250.
Hayes, B. (2007) Trains of thought. American Sci¬
entist, 95 (2], 108-113.
Hazlett, H. C. (2005] Klinefelter Syndrome. In
S. Goldstein & C. R. Reynolds (Eds.], Hand¬
book of neurodevelopmental and genetic orders
in adults. New York: Guilford Press, pp. 367—
382.
Hearnshaw, L. S. (1979) Cyril Burt, psychologist.
London: Hodder & Stoughton.
Hedges, L. V., & Nowell, A. (1995) Sex differ¬
ences in mental test scores, variability, and
numbers of high-scoring individuals. Science,
270, 365-367.
Hedges, L. V., & Nowell, A. (1998) Black- White
test score convergence since 1965. In C. Jenks
& M. Phillips (Eds.], The black-white test score
gap. Washington: Brookings Institute, 129-181.
Hedlund, J., Forsythe, G. B., Horvath, J. A.,
Williams, W. M., Snook, S., & Sternberg, R.
J. (2003) Identifying and assessing tacit knowl¬
edge: Understanding the practical intelligence
of military leaders. The Leadership Quarterly,
14, 117- 140.
Hedlund, J., Wilt, J. M., Nebel, K. L., Ashford,
S. J., & Sternberg, R. J. (2006) Assessing practi¬
cal intelligence in business school admissions:
A supplement to the graduate admissions test.
Learning and Individual Differences, 16, 101-127.
Hegarty, M., Just, M. A., & Morrison, I. R. (1988)
Mental models of mechanical systems: Indi¬
vidual differences in qualitative and quanti¬
tative reasoning. Cognitive Psychology, 20, 191-
236.
Hegarty, M., & Waller, D. A. (2005) Individual
differences in spatial abilities. In P. Shah &
A. Miyake (Eds.], The Cambridge handbook of
visuospatial thinking. New York: Cambridge
University Press, pp. 121-169.
Heitz, R. P., Unsworth, N., & Engle, R. W. (2005]
Working memory capacity, attention control,
and fluid intelligence. In O. Wilhelm & R. W.
Engle (Eds.], Handbook of understanding and
measuring intelligence. Thousand Oaks, CA:
Sage, pp. 61-78.
Helms-Lorenz, M., Van de Vijver, F. J. R., &
Poortinga, Y. H. (2003) Cross-cultural differ¬
ences in cognitive performance and Spear¬
man's hypothesis: g or c? Intelligence, 31 (1],
9-29.
Herrnstein, R. J., 81 Murray, C. (1994] The Bell
Curve: Intelligence and class structure in Amer¬
ican life. New York: The Free Press.
Hick, W. E. (1952) On the rate of gain of infor¬
mation. Quarterly Journal of Experimental Psy¬
chology, 4, 11-26.
Hill, L., Chorney, M. J., Lubinski, D., Thompson,
L. A., & Plomin, R. (2002] A quantitative trait
locus not associated with cognitive ability in
children: A failure to replicate. Psychological
Science, 13 (6), 561-562.
Hill, N., & Schneider, W. (2006] Brain changes
and the development of expertise: Neu-
roanatomical and neurophysiological evidence
about skill-based adaptations. In K. A. Eric¬
sson, N. Charness, P. J. Feltovich, & R. R.
Hoffman, The Cambridge handbook of exper¬
tise and expert performance. Cambridge: Cam¬
bridge University Press, pp. 653-682.
Hines, M. (2007) Do sex differences in cognition
cause the shortage of women in science? In
S. J. Ceci & W. M. Williams (Eds.], Why aren't
there more women in science: Top researchers
debate the evidence. Washington, DC: Ameri¬
can Psychological Association, 101-112.
Hoeft, F., Ueno, T., Reiss, A. L., Meyler, A.,
Whitfield-Gabrieli, S., Glover, G. H., Keller,
T. A., Kobayashi, N., Mazaika, P., Jo, B., Just,
M. A., & Gabrieli, J. D.E. (2007] Prediction
of children's reading skills using behavioral,
functional, and structural neuroimaging mea¬
sures. Behavioral Neuroscience, 121 (3], 602-613.
466
REFERENCES
Holohan, C. K., & Sears, R. R. (1995) The gifted
group in later maturity. Stanford, C A: Stanford
University Press.
Horn, J. L. (1985) Remodeling old models of intel¬
ligence. In B. B. Wolman (Ed.), Handbook of
intelligence: Theories , measurements, and appli¬
cations. New York: Wiley, pp. 267-300.
Horn, J. L. (1986) Intellectual ability concepts. In
R. L. Sternberg (Ed.), Advances in the psychol¬
ogy of human intelligence, Vol 3. Hillsdale, NJ:
Erlbaum, pp. 35-78.
Horn, J. L. (1998) A basis for research on age
differences in cognitive capabilities. In J. J.
McArdle & R. W. Woodcock (Eds.), Human
cognitive abilities in theory and practice. Mah-
wah, NJ: Erlbaum, pp. 57-91.
Horn J. L. & Noll, J. (1994) A system for
understanding cognitive capabilities: A the¬
ory and the evidence on which it is based. In
D. Detterman (Ed.), Current topics in human
intelligence: Vol. 4. Theories of intelligence.
Norwood, NJ: Ablex, pp. 151-204.
Horn, J. L., & Stankov, L. (1982) Auditory and
visual factors in intelligence. Intelligence, 6 (2),
165-185.
Howard, A., & Bray, D. W. (1988) Managerial
lives in transition: Advancing age and changing
times. New York: Guilford Press.
Hoyer, W. J., & Verhaeghen, P. (2006) Memory
aging. In J. W. Birren & K. W. Schaie (Eds.),
Handbook of the psychology of aging (6th edi¬
tion). Amsterdam: Elsevier, 209-232.
Huarte de San Juan, J. (1575/1991) Examen de
ingenios para las ciencias (edicion Felisa Fresco
Otero). Madrid: Espasa-Calpa S. A.
Hubbs-Tait, L., Nation, J. R., Krebs, N. F.,
& Bellinger, D. C. (2005) Neurotoxicants,
micronutrients, and social environments: Indi¬
vidual and combined effects on children's
development. Psychological Science in the Pub¬
lic Interest, 6 (3), 57-121.
Humphreys, L. G., & Lubinski, D. J. (1996)
Assessing spatial visualization: An underap¬
preciated ability for many school and work
settings. In C. P. Benbow & D. J. Lubin-
ski (Eds.), Intellectual talent: Psychometric and
social issues. Baltimore, MD: Johns Hopkins
University Press, pp. 116-140.
Hunt, E. (1980) Intelligence as an information
processing concept. British Journal of Psychol-
°&, 71 (4), 449-474-
Hunt, E. (1987} The next word on verbal ability.
In P. A. Vernon (Ed.), Speed of information
processing and intelligence. New York: Ablex,
PP- 347“392-
Hunt, E. (1995a) The role of intelligence in
modern society. American Scientist, 83 (July-
August), 356-368.
Hunt, E. (1995b) Will we be smart enough? A cogni¬
tive analysis of the coming workforce. New York:
Russell Sage.
Hunt, E. (1997) Nature vs. nurture: The feeling
of vuja de. In R. J. Sternberg & E. Grigorenko
(Eds.), Intelligence, heredity, and environment.
Cambridge: Cambridge University Press,
pp. 531-551.
Hunt, E. (2002) Thoughts on thought. Mahwah,
NJ: Erlbaum.
Hunt, E. (2004) Multiple views of multiple intel¬
ligence. Contemporary Psychology, 46 (1), 5-
7*
Hunt, E. (2007) The mathematics of behavior.
Cambridge: Cambridge University Press.
Hunt, E. (2008) Applying the theory of successful
intelligence to education - The good, the bad,
and the ogre: Commentary on Sternberg et al.
(2008). Perspectives on Psychological Science, 3
(6), 509-515.
Hunt, E., & Agnoli, F. (1991) The Whorfian
hypothesis: A cognitive psychology perspec¬
tive. Psychological Review, 98 (3), 377-389.
Hunt, E., & Carlson, J. (2007a) Considerations
relating to the study of group differences
in intelligence. Perspectives in Psychological
Science, 2 (2), 194-213.
Hunt, E., & Carlson, J. (2007b) Reply to com¬
mentators. Perspectives in Psychological Science,
2 (2), 224-226.
Hunt, E., & Carlson, J. (2007c) The standards
for conducting research on topics of immedi¬
ate social relevance. Intelligence, 35 (5), 393-
399-
Hunt, E., Frost, N., & Lunneborg, C. E. (1973)
Individual differences in cognition: A new
approach to intelligence. In G. S. Bower
(Ed.), Advances in learning and motivation
(vol. 7). New York: Academic Press, pp. 87-
123-
Hunt, E., & Lansman, M. (1986) A unified model
of attention and problem solving. Psychobgical
Review, 93, 446-461.
Hunt, E., & Lunneborg, C. E., & Lewis, J. (1975)
What does it mean to be high verbal? Cognitive
Psychology, 7, 194-227.
Hunt, E., & Madhyastha, T. M. (2008) Recruit¬
ment modeling: An analysis and application to
the study of male-female differences in intel¬
ligence. Intelligence, 36 (6), 653-663.
Hunt, E., & Minstrell, J. (1994) A cognitive
approach to the teaching of physics. In K.
REFERENCES
467
McGilly (Ed.), Classroom lessons: Integrating
cognitive theory in the classroom. Cambridge,
MA: MIT Press, pp. 51-74.
Hunt E., & Minstrell, J. (1996] Effective instruc¬
tion in science and mathematics: Psycholog¬
ical principles, social constraints. Issues in
Education: Contributions from Educational Psy¬
chology, 2 (2), 123-162.
Hunt, E., Pellegrino, J. W., Frick, R. W., Farr,
S., & Alderton, D. (1988] The ability to rea¬
son about movement in the visual field. Intel¬
ligence, 12, 77-100.
Hunt, E., & Wittmann, W. (2008) National intel¬
ligence and national prosperity. Intelligence, 36
0), 1-9-
Hunt, S. [2007) Hiring for success: The art and sci¬
ence of staffing assessment and employee selec¬
tion. San Francisco: Pfieffer.
Hunter, J. E., & Schmidt, F. L. (1990) Methods
of meta- analysis: Correcting error and bias in
scientific findings. Newbury Park, CA: Sage.
Hutchins, E. (1983] Understanding micronesian
navigation. In D. Gentner & A. L. Stevens
(Eds.), Mental models. Hillsdale, NJ: Erlbaum,
pp. 191—226.
Hyde, J. S. (2007) Women in science: Gen¬
der similarities in abilities and sociocultural
forces. In S. J. Ceci & W. M. Williams
(Eds.), Why aren't there more women in science:
Top researchers debate the evidence. Washing¬
ton, DC: American Psychological Association,
pp. 131-145.
Hyde, J. S., Lindberg, S. M., Linn, M. C., Ellis, A.
B., & Williams, C. C. (2008) Gender similar¬
ities characterize math performance. Science,
321 (25 July), 494-495.
Hyman, R. (1953) Stimulus information as
a determinant of reaction-time. Journal of
Experimental Psychology, 45, 188-196.
Inlow, J. K., & Restifo, L. L. (2004) Molecular
and comparative genetics of mental retarda¬
tion. Genetics, 166 (2), 835-881.
Irwing, P., Hamza, A., Khaleefa, O., & Lynn, R.
(2008) Effects of abacus training on the intel¬
ligence of Sudanese children. Personality and
Indiindual Differences, 45 (7), 694-696.
Irwing, P., & Lynn, R. (2005) Sex differences
in means and variabilities on the progres¬
sive matrices in university students: A meta¬
analysis. British Journal of Psychology, 96, 505-
524.
Isaacson, W. (2003) Benjamin Franklin: An Amer¬
ican life. New York: Simon and Schuster.
Jackson, D. N., & Rushton, J. P. (2006) Males
have greater g: Sex differences in general
mental ability from 100,000 17- to 18 -year-olds
on the Scholastic Assessment Test. Intelligence,
34 (5). 479-486.
Jacoby, L. L., Toth, J. P., & Yonelinas, A. P. (1993)
Separating conscious and unconscious influ¬
ences on memory: Measuring recollections.
Journal of Experimental Psychology: General,
122 (2), 139-154.
Jaeggi, S. M., Buschkeuhl, M., Jonides, J., & Per-
rig, W. J. (2008) Improving fluid intelligence
with training on working memory. Proceedings
of the National Academy of Science, 105 (19),
6829-6833.
Jarvik, L. F., Klodin, V., & Matsuyama, S. S.
(1973) Human aggression and the extra Y chro¬
mosome: Fact or fantasy? American Psycholo¬
gist, 28 (8), 574-682.
Jensen, A. R. (1969) How much can we boost IQ
and scholastic achievement? Harvard Educa¬
tional Remew, 39 (1), 1-123.
Jensen, A. R. (1974) Kinship correlations reported
by Sir Cyril Burt. Behavior Genetics, 4 (1), 1-28.
Jensen, A. R. (1980) Bias in mental testing. New
York: Free Press.
Jensen, A. R. (1993) Spearman’s hypothe¬
sis tested with chronometric information¬
processing tasks. Intelligence, 17 (1), 47-77.
Jensen, A. R. (1997) The puzzle of non-genetic
variance. In R. J. Sternberg & E. Grigorenko
(Eds.), Intelligence, heredity, and environment.
Cambridge: Cambridge University Press,
pp. 42-88.
Jensen, A. R. (1998) The g factor: The science of
mental ability. Westport, CT: Praeger Publish¬
ers/Greenwood Publishing Group.
Jensen, A. R. (2002) Psychometric g: Definition
and substantiation. In R. J. Sternberg & E.
L. Grigorenko (Eds.), The general factor of
intelligence: How general is it? Mahwah, NJ:
Erlbaum, pp. 39-53.
Jensen, A. R. (2006) Cbcking the mind: Mental
chronometry and individual differences: Ams¬
terdam: Elsevier.
Johnson, W. (2007) Genetic and environmen¬
tal influences on behavior: Capturing all the
interplay. Psychological Remew, 114 (2), 423-
440.
Johnson, W., 8l Bouchard, T. J., Jr. (2005a) The
structure of human intelligence: It is verbal,
perceptual, and image rotation (VPR), not
fluid and crystallized. Intelligence, 33 (4), 393-
416.
Johnson, W., & Bouchard, T. J. Jr. (2005b) Con¬
structive replication of the visual perceptual
image-rotation model in Thurstone’s (1941)
468
REFERENCES
battery of 60 tests of mental ability. Intelli¬
gence, 33 (4), 417-43°.
Johnson, W., & Bouchard, T. J., Jr. (2007a) Sex
differences in mental abilities: g masks the
dimensions on which they lie. Intelligence, 35
(0< 23-39-
Johnson, W., & Bouchard, T. J., Jr. (2007b)
Sex differences in mental ability: A proposed
means to link them to brain structure and
functioning. Intelligence, 35 (3), 197-209.
Johnson, W., Bouchard, T. J., Jr., Krueger, R. F.,
McGue, M., & Gottesman, 1. 1. (2004) Just one
g: Consistent results from three test batteries.
Intelligence, 32, 95-107.
Johnson, W., Bouchard, T. J., Jr., McGue, M.,
Segal, N. L., Tellegen, A., Keyes, M., &
Gottesman, I. I. (2007) Genetic and environ¬
mental influences on the Verbal-Perceptual-
Image Rotation (VPR) model of the structure
of mental abilities in the Minnesota study of
twins reared apart. Intelligence, 35 (6), 542-562.
Johnson, W., Carothers, A., & Deary, I. J. (2008)
Sex differences in variability in general intelli¬
gence: A new look at an old question. Perspec¬
tives on Psychological Science, 3 (6), 518-531.
Johnson, W., Carothers, A., & Deary, I. J. (2009)
A role for the X chromosome in sex differ¬
ences in variability in general intelligence. Per¬
spectives on Psychological Science, 4 (6), 598-611.
Johnson, W., te Nijenhuis, J., & Bouchard,
T. J., Jr. (2007) Replication of the hierarchical
visual-perception-rotation model in De Wolff
and Buitens (1963) battery of 46 tests of men¬
tal ability. Intelligence, 35 (3), 69-81.
Johnson-Laird, P. N., & Byrne, M. J. B. (1991)
Deduction. Hillsdale, NJ: Erlbaum.
Jones, B. M. (1971) Verbal and spatial intelligence
in short and long term alcoholics. Journal of
Nervous and Mental Disease, 154 (4), 292-297.
Jones, S. M., & Zigler, E. (2002) The Mozart
effect: Not learning from history. Journal of
Applied Developmental Psychology , 23 (3), 355—
372-
Joreskog, K. G., & Sorboom, D. (1979) Advances
in factor analysis and structural equation mod¬
eling. Cambridge, MA: Abt Books.
Juan-Espinosa, M., Garcia, L. F., Escorias, S.,
Rebollo, I., Colom, R., & Abad, F. J. (2002)
Age de differentiation hypothesis: Evidence
from the WAIS III. Intelligence, 30 (5), 395-
408.
Jung, R. E., & Haier, R. J. (2007) The parieto¬
frontal integration theory (P-FIT) of intel¬
ligence: Converging neuroimaging evidence.
Behavioral and Brain Sciences, 30, 135-187.
Just, M. A., & Carpenter, P. A. (1985) Cogni¬
tive coordinate systems: Accounts of mental
rotation and individual differences in spatial
ability. Psychological Review, 92 (2), 137-172.
Just, M. A., & Carpenter, P. A. (1992) A capac¬
ity theory of comprehension: Individual dif¬
ferences in working memory. Psychological
Review, 99, 122-149.
Just, M. A., Carpenter, P. A., & Keller, T. A.
(1996) The capacity theory of comprehension:
New frontiers of evidence and arguments. Psy¬
chological Review, 104 (4), 773-780.
Just, M. A., Carpenter, P. A., Keller, T. A., Eddy,
W. A., Martijn, F., van Dijl, J. M., Suda, K.p
Schatz, G. et al. (1996) Brain activation modu¬
lated by sentence comprehension. Science, 274
(5284), 114-116.
Kanaya, T., Scullin, M. H., & Ceci., S. J. (2003)
The Flynn Effect and U. S. policy: The impact
of rising IQ scores on American society via
mental retardation diagnoses. American Psy¬
chologist, 58 (10), 778-790.
Kanazawa, S. (2004) General intelligence as
a domain-specific adaptation. Psychological
Review, 111 (2), 512-523.
Kanazawa, S. (2008) Temperature and evolution¬
ary novelty as forces behind the evolution of
general intelligence. Intelligence, 36 (2), 99-108.
Kane, M. J., Brown, L. H., McVay, J. C., Silvia,
P. J., Myin-Germeys, I., & Kwapil, T. R. (2007)
For whom the mind wanders, and when: An
experience-sampling study of working mem¬
ory and executive control in daily life. Psycho¬
logical Science, 18 (7), 614—621.
Kane, M. J., & Engle, R. W. (2002) The role of pre¬
frontal cortex in working memory capacity,
executive attention, and general fluid intelli¬
gence: An individual differences perspective.
Psychonomic Bulletin and Review, 9 (4), 637—
671.
Kane, M. J., Hambrick, D. Z., Conway, A. R. A.,
& Engle, R. W. (2001) The generality of
working memory capacity: A latent variable
approach. Journal of Experimental Psychology:
General, 130, 169-183.
Kane, M. J., Hambrick, D. Z., Tuholski, S. W.,
Wilhelm, O., Payne, T. W., & Engle, R. W.
(2004) The generality of working memory
capacity: A latent variable approach to verbal
and visual-spatial working memory span and
reasoning. Journal of Experimental Psychology:
General , 133, 189-217.
Kaplan, C. (1996) Predictive validity of the
WPPSI-R: A four year follow-up study. Psy¬
chology in the Schools, 33 (3), 211-220.
REFERENCES
469
Karama, S., Ad-Dab’bagh, Y., Haier, R. J., Deary,
I. J., Lyttelton, O. C., Lepage, C., & Evans,
A. C. (2009) The Brain Development Co¬
operative Group: Positive association between
cognitive ability and cortical thickness in a
representative US sample of 6 to 18 year olds.
Intelligence , 37 (2), 145-155.
Kaufman, A. S. (2001) Do low levels of lead pro¬
duce IQ loss in children? A careful examina¬
tion of the literature. Archives of Clinical Neu¬
ropsychology, 16, 303-341.
Kimura, D. [1999) Sex and cognition. Cambridge,
MA: MIT Press.
Kintsch, W. (1998) Comprehension: A paradigm
for cognition. New York: Cambridge University
Press.
Kirschner, P. A., Sweller, J., & Clark, R. E.
(2006) Why minimal guidance during instruc¬
tion does not work: An analysis of the failure
of constructivist, discovery, problem-based,
experiential, and inquiry-based teaching. Edu¬
cational Psychologist, 41, 75-86.
Klein, G. (1998) Sources of power: How people
make decisions. Cambridge, MA: MIT Press.
Klein, G. (2009) Streetlights and shadows. Cam¬
bridge, MA: MIT Press.
Kobrin, J. L., Patterson, B. F., Shaw, E. J., Mat-
tern, K. D., & Barbuti, S. M. Validity of the
SAT for predicting first year college grade
point average. College Board Research Report
2008-5. New York: The College Board.
Koenig, K. A., Frey, M. C., & Detterman,
D. K. (2008) ACT and general cognitive abil¬
ity. Intelligence, 36 (2), 153-160.
Kornhauser, A. W. (1918) The economic standing
of parents and the intelligence of their chil¬
dren. Journal of Educational Psychology, 9 (3),
159-164.
Kosslyn, S. M. (1980) Image and mind. Cam¬
bridge, MA: Harvard University Press.
Kosslyn, S. M. (1994) Image and brain. Cam¬
bridge, MA: MIT Press.
Kotchick, B. A., & Forehand, R. (2002) Putting
parenting in perspective: A discussion of the
contextual factors that shape parenting prac¬
tices. Journal of Child and Family Studies, 11
(3). 255-2^9-
Kovarik, W. (2005) Ethyl leaded gasoline: How a
classic occupational disease became an inter¬
national public health disaster. International
Journal of Occupational Health, 11 (4), 384-397.
Kovas, Y., Haworth, C. M. A., Dale, P. S., &
Plomin, R. [2007) The genetic and environ¬
mental origins of learning abilities and disabil¬
ities in the early school years. Monographs of
the Society for Research in Child Development,
72 (3), vii-160.
Kramer, A. F., Fabiani, M., & Colcombe, S. J.
(2006) Contributions of cognitive neuro¬
science to the understanding of behavior and
aging. In J. E. Birren & K. W. Schaie (Eds.),
Handbook of the psychology of aging, 6th edi¬
tion. Burlington, MA: Elsevier, 57-83.
Kremen, W. S., Jacobsen, K. C., Xian, H., Eisen,
S. A., Eaves, L. J., Tsuang, M. T., & Lyons, M.
J. (2007) Genetics of verbal working memory
processes: A twin study of middle-aged men.
Neuropsychology, 21 (5), 569-580.
Kristensen, P., & Bjerkedal, T. (2007) Explaining
the relation between birth order and intelli¬
gence. Science, 316 (5832), 1717.
Kroc, R., Howard, R., Hull, P., & Woodard,
D. (1997) Graduation rates: Do students’
academic program choices make a difference?
Paper presented at the 37th Annual Forum
of the Association for Institutional Research
(Orlando, FL, May 1997). Dowloaded from
aer.arizona.edu/enrollment%20management/
whitepapersZgraduation_rates.htm, Decem¬
ber 2008.
Kronman, A. T. (2007) Against political correct¬
ness: A liberal’s cri de coeur. Yale Alumni
Masgazine, 71 (1), September-October.
Kuncel, N. R.; & Hezlett, S. A. (2007) Standard¬
ized tests predict graduate students’ success.
Science, 315 (23 February), 1080-1081.
Kuncel, N. R., Hezlett, S. A., & Ones, D. S.
(2004) Academic performance, career poten¬
tial, creativity, and job performance: Can one
construct predict them all? Journal of Per¬
sonality and Social Psychology, 86 (1), 148-
161.
Kyllonen, P. C. (1993) Aptitude testing inspired
by information processing: A test of the four
sources model. Journal of General Psychology,
120 (3). 375“4° 5-
Kyllonen, P. C., & Christal, R. E. (1990) Reason¬
ing ability is (little more than) working mem¬
ory capacity?! Intelligence, 14, 389-433.
Kyllonen, P. C., & Stephens, D. L. (1990) Cog¬
nitive abilities as determinants of success in
acquiring logic skills. Learning and Individual
Differences, 2 (2), 129-160.
Lai, C. S. L., Fisher, S. E., Hurst, J. A., Vargha-
Khadem, F., & Monaco, A. P. (2001) A
forkhead-domain gene is mutated in a severe
speech and language disorder. Nature, 413
(6855), 519-523.
Lam, L. T., & Kirby, S. (2002) Is emotional
intelligence an advantage? An exploration of
470
REFERENCES
emotional intelligence. The Journal of Social
Psychology, 142 (1), 133-143.
Lange, R. T., Chelune, G. J., Taylor, M. J.,
Woodward, T. S., & Heaton, R. K. (2006)
Development of demographic norms for four
new WAIS-IIIAVMS-III indexes. Psychological
Assessment, 19 (2), 174-181.
Lanphear, B. P., Khoury, J., Yolton, K., Baghurst,
P., and eleven further authors. (2005) Low-
level environmental lead exposure and chil¬
dren’s intellectual function: An international
pooled analysis. Environmental Health Perspec¬
tives, 113 (7), 894-899.
Larson, G. E., Haier, R. J., LaCasse, L., & Hazen,
K. (1995) Evaluation of a “mental effort”
hypothesis for correlations between cortical
metabolism and intelligence. Intelligence, 21
(3), 267-278.
Larson, G. E., & Saccuzzo, D. P. [1989) Cognitive
correlates of intelligence: Toward a process
theory of g. Intelligence, 13 (1), 5-31.
Lave, J. (1988] Cognition in practice. Cambridge:
Cambridge University Press.
Law, D. J., Pellegrino, J. W., & Hunt, E. B. (1993)
Comparing the tortoise and the hare: Gender
differences and experience in dynamic spatial
reasoning tasks. Psychological Science, 4 (1), 35-
4°.
Leckliter, I. N., & Matarazzo, J. D. (1989] The
influence of age, education, IQ, gender, and
alcohol abuse on the Halstead-Reitan Neu¬
ropsychological Test Battery performance.
Journal of Clinical Psychology, 145 (4}, 484-
512.
Lee, H. L., Devlin, J. T., Shakeshaft, C., Stewart,
L. H., Brennan, A., Glensman, J., Pitcher, K.,
Crinion, J., Mechelli, A., Frackowiak, R. S. J.,
Green, D. W., & Price, C. J. (2007) Anatomi¬
cal traces of vocabulary acquisition in the ado¬
lescent brain. Journal of Neuroscience, 27 (5),
1184-1189.
Legree, P. J., Pifer, M. E., & Grafton, F. C.
(1996) Correlations among cognitive abilities
are lower for higher ability groups. Intelligence,
23 M> 45“57-
Lemann, N. (1999) The big test: The secret his¬
tory of American meritocracy. New York: Far¬
rar, Strous, and Giroux.
Lesser, G. S., Fifer, G., & Clark, D. H. (1965)
Mental abilities of children from different
social-class and cultural groups. Monographs of
the Society for Research in Child Development,
30 [4), 1-115.
Levidow, B. (1993) The effect of high school
physics instruction on measures of general
knowledge and reasoning ability. Unpublished
Ph.D. dissertation, University of Washington.
Lewontin R. C. (1974) The genetic basis of evolu¬
tionary change. New York: Columbia Univer¬
sity Press.
Lidz, C., & Elliot, J. (Eds.) (2000) Dynamic assess¬
ment. New York: JAI Press.
Lindzey, G., Loehlin, J. C., & Spuhler, J. (1975)
Racial differences in intelligence. San Francisco:
Freeman.
Lippmann, W. (1922a) The mental age of Ameri¬
cans. New Republic, 32 (412) (25 October), 213-
215.
Lippmann, W. (1922b) The mental age of Amer¬
icans. New Republic, 32 (413) (1 November),
246-248.
Loehlin, J. C. (1998) Wither dysgenics? Comment
on Lynn and Preston. In U. Neisser (Ed.),
The rising curve: Long-term gains in IQ and
related measures. Washington, DC: American
Psychological Association, 389-398.
Loehlin, J. C. (2004) Latent variable models: An
introduction to factor, path, and structural equa¬
tion analysis. Mahwah, NJ: Erlbaum.
Loehlin, J. C., Horn, J. M., & Ernst, J. L.
(2007) Genetic and environmental influences
on adult life outcomes: Evidence from the
Texas Adoption Project. Behavior Genetics, 37
(3), 463-476.
Loehlin, J. C., Horn, J. M., & Willerman, L.
(1997) The Texas Adoption Project. In R. J.
Sternberg E. Grigorenko (Eds.), Intelligence,
heredity, and environment. Cambridge: Cam¬
bridge University Press, pp. 105-125.
Logie, R. H. (1995) Visuo-spatial working memory.
Hove, UK: Erlbaum.
Lohman, D., & Hagen, E. P. (2001) Cognitive abil¬
ities test: Form 6. Chicago: Riverside Press.
Longman, R. S., Sakofske, D. H., & Fung, T. S.
(2007) WAIS-III percentile scores by educa¬
tion and sex for U. S. and Canadian popula¬
tions. Assessment, 14 (4), 426-432.
Lopez, A., Atran, S., Coley, J. D., Medin, D. L.,
& Smith, E. E. (1997) The tree of life: Uni¬
versal and cultural features of folkbiological
taxonomies and inductions. Cognitive Psychol¬
ogy , 32 (3), 251-295.
Lord, F. M. (1956) A study of speed factors in
tests and academic grades. Psychometrika, 52
(6), 505-510.
Lord, F. M., & Novick, M. R. (1968) Statistical
theories of mental test scores. Reading, MA:
Addison-Wesley.
Lubinski, D., & Benbow, C. P. (2006) Study
of mathematically precocious youth after
REFERENCES
471
35 years: Uncovering antecedents for the
development of math -science expertise. Per¬
spectives on Psychological Science, 1 (4), 316-345.
Lubinski, D.; Benbow, C. P., Webb, R. M.,
& Bleske-Rechek, A. (2006) Tracking excep¬
tional human capital over two decades. Psy¬
chological Science , 17 (3), 194-199.
Luciano, M., Smith, G. A., Wright, M. J.,
Geffen, G. ML, Geffen, L. B.; & Martin, N.
(2001) On the heritability of inspection time
and its covariance with IQ: A twin study. Intel¬
ligence, 29 (6), 443-457.
Luciano, M., Wright, M. J., Duffy, D. L., Wain-
wright, M. A., Zhu, G., Evans, D. M., Geffen,
G. M., Montgomery, G. W., & Martin, N. G.
(2006) Genome-wide scan of IQ finds signifi¬
cant linkage to a quantitative trait locus on 2q.
Behavior Genetics, 36 (1), 45—55.
Luciano, M., Wright, M. J., Smith, G. A., Gef¬
fen, G. M., Geffen, L. B., & Martini, N. G.
(2001) Genetic covariance among measures of
information processing speed, working mem¬
ory, and IQ. Behavior Genetics, 31 (6), 581-592.
Luders, E., Narr, K. L., Thompson, P. M., &Toga,
A. W. (2009) Neuroanatomical correlates of
intelligence. Intelligence, 37 (2], 156-163.
Luria, A. R. (1962/1980) Higher cortical functions in
man (second edition, revised and expanded).
New York: Basic Books. (English version, orig¬
inal in Russian.)
Lynn, R. (1990) The role of nutrition in secular
increase in intelligence. Personality and Indi¬
vidual Differences, 11 (3), 273-285.
Lynn, R. (1994) Sex differences in intelligence and
brain size: A paradox resolved. Personality and
Individual Differences, 17, 257-271.
Lynn, R. (1998) The decline of genotypic intel¬
ligence. In U. Neisser (Ed.), The rising curve:
Long-term gains in IQ and related measures.
Washington, DC: American Psychological
Association, pp. 333-354.
Lynn, R. (1999) Sex differences in intelligence
and brain size: A developmental theory. Intel¬
ligence, 27 (1), 1-12.
Lynn, R. (2006) Race differences in intelligence: An
evolutionary analysis. Augusta, GA: Washing¬
ton Summit Publishers.
Lynn, R., Allik, J., & Irwing, P. (2004) Sex dif¬
ferences on three factors identified in Raven's
Standard Progressive Matrices. Intelligence, 32
(4), 411-424.
Lynn, R., Backhoff, E., & Contreras, L. A. (2005)
Ethnic and racial differences on the standard
progressive matrices in Mexico. Journal of
Biosocial Sciences, 37, 107-113.
Lynn, R., & Harvey, J. (2008) The decline of the
world’s IQ. Intelligence, 36 (2), 112-120.
Lynn, R., & Hattori, K. (1990) The heritability
of intelligence in Japan. Behairior Genetics, 20
(4), 545“ 54^-
Lynn, R., & Irwing, P. (2004a) Sex differences
on the progressive matrices: A meta-analysis.
Intelligence, 32 (5), 481-498.
Lynn, R., & Irwing, P. (2004b) Sex differences
on the advanced progressive matrices in col¬
lege students. Personality and Individual Dif¬
ferences, 37 (1), 219-223.
Lynn, R., & Irwing, P. (2005) Sex differences
in means and variability on the progres¬
sive matrices in university students: A meta¬
analysis. British Journal of Psychology, 96, 505-
524.
Lynn, R., & Mau, W.-C. (2001) Ethnic and
sex differences in the predictive validity of
the Scholastic Achievement Test for college
grades. Psychological Reports, 88 (3, pt. 2),
1099-1104.
Lynn, R., & Meisenberg, G. (2010) The average
IQ of sub-Saharan Africans: Comments on
Wicherts, Dolan, and van der Maas. Intelli¬
gence, 38 (1), 21-29.
Lynn, R., & Owen, K. (1994) Spearman’s hypoth¬
esis and test score differences between White,
Indian, and Blacks in South Africa. The Jour¬
nal of General Psychology, 121 (1), 27-36.
Lynn, R., & Van Court, M. (2004) New evi¬
dence of dysgenic fertility for intelligence in
the United States. Intelligence, 32 (2), 193—
201.
Lynn, R., & Vanhanen, T. (2002) IQ and the
wealth of nations. Westport, CT: Praeger.
Lynn, R., & Vanhanen, T. (2006) IQ and global
inequality. Augusta, GA: Washington Summit
Publishers.
MacDonald, M. C., Just, M. A., & Carpenter,
P. A. (1992) Working memory constraints on
the processing of syntactic ambiguity. Cogni¬
tive Psychology, 24, 56-98.
Mace, R. (2008) Reproducing in cities. Science,
319 (8 February), 764-766.
Machin, S., & Pekkarinen, T. (2008) Global sex
differences in test score variability. Science, 322
(28 November), 1331-1332.
Macintosh, N. J. (1995) Cyril Burt: Fraud or
framed? Oxford: Oxford University Press.
Macintosh, N. J. (1998) IQ and human intelligence.
Oxford: Oxford University Press.
MacLeod, C. M., Hunt, E., & Mathews, N. N.
(1978) Individual differences in the verifica¬
tion of sentence-picture relationships. Journal
REFERENCES
472
of Verbal Learning and Verbal Behavior , 17,
493-507.
Madhyastha; T. M., Hunt, E., Deary, I. J., Gale,
C. R.; & Dykiert, D. (2009) Recruitment mod¬
eling applied to longitudinal studies of group
differences in intelligence. Intelligence, 37 [4],
422-427.
Maguire, E. A., Burgess, N., & O’Keefe, J. (1999)
Human spatial navigation: Cognitive maps,
sexual dimorphism, and neural substrates.
Current Opinion in Neurobiology, 9 (2), 171—
177.
Maguire, E. A., Spiers, H. J., Good, C. D., Hart¬
ley, T., Frackowiak, R. S., & Burgess, N. (2003)
Navigation expertise and the human hip¬
pocampus: A structural brain imaging anal¬
ysis. Hippocampus, 13, 250-259.
Malinowski, J. C. (2001] Mental rotation and real-
world wayfinding. Perceptual and Motor Skills,
92 (Oi 19_3°-
Marsiske, M., & Margrett, J. A. (2006) Everyday
problem solving and decision making. In J. E.
Birren & K. W. Schaie (Eds.), Handbook of the
Psychology of Aging (6th edition). Burlington,
MA: Elsevier, 315-342.
Marvis, C. B. (2007) Language and commu¬
nicative development in Williams' syndrome.
Mental Retardation and Developmental Dis¬
abilities Research Reviews, 13 (1), 3-15.
Matarazzo, J. D. (1972) Wechsler’s measurement
and appraisal of adult intelligence. Baltimore:
Williams and Wilkins.
Matarazzo, J. D., Bornstein, R. A., McDermott,
P. A., & Noonan, J. V. (1986) Verbal IQ vs.
performance IQ difference scores in males
and females from the WAIS-R standardiza¬
tion sample. Journal of Clinical Psychology, 42
(6), 965-974.
Mathews, N. N., Hunt, E., & MacLeod, C. M.
(1980) Strategy choice and strategy training in
sentence-picture verification. Journal of Verbal
Learning and Verbal Behavior, 19, 531-538.
Matte, T. D., Bresnahan, M., Begg, M. D., &
Susser, E. (2001) Influence of variation in birth
weight within normal range and within sib-
ships on IQ at age 7 years: Cohort study.
British Medical Journal , 323 (7308), 310-314.
Mattern, K. D., Pattern, B. F., Shaw, E. J.,
Kobrin, J. F., & Barbuti, S. M. (2008) Dif¬
ferential validity and prediction of the SAT.
College Board Research Report 2008-4. New
York: College Entrance Examination Board.
Matthews, G., Zeidner, M., & Roberts, R. D.
(2005) Emotional intelligence: An elusive abil¬
ity? In O. Wilhelm & R. W. Engle (Eds.),
Handbook of understanding and measuring
intelligence. Thousand Oaks, CA: Sage, pp. 79-
99-
Mau, W.-C., & Lynn, R. (2000) Gender differ¬
ences in homework and test scores in math¬
ematics, reading and science at tenth and
twelfth grade. Psychology , Evolution, Gender,
2 (2), 119-125.
Mau, W.-C., Lynn, R. (2001) Gender differences
on the Scholastic Aptitude Test, the American
College Test, and college grades. Educational
Psychology, 21 (2), 133-136.
Mauger, P. A., Kolmodin, C. A. (1975) Long-term
predictive validity of the scholastic aptitude
test. Journal of Educational Psychology, 67 (6),
847-851.
Max, J. E., Manes, F. F., Robertson, B. A. M.,
Mathews, K., Fox, P. T., & Lancaster, J. (2005)
Prefrontal and executive attention network
lesions and the development of attention-
deficit/hyperactivity symptomatology. Jour¬
nal of the American Academy of Child and Ado¬
lescent Psychiatry, 44 (5), 443-450.
Mayer, J. D., Salovey, P., Caruso, D. R. (2004)
Emotional intelligence: Theory, findings, and
implications. Psychological Inquiry, 15 (3), 197-
215-
Mayer, R. E. (2002) Should there be a three -
strikes rules against pure discovery learning?
The case for guided methods of instruction.
American Psychologist, 59 (1), 14-19.
McArdle, J. J. (1998) Contemporary statistical
models for examining test-bias. In J. J. McCar-
dle & R. W. Woodcock (Eds.), Human cogni¬
tive abilities: Theory and practice. Mahwah, NJ:
Erlbaum, pp. 157-196.
McArdle, J. J., Ferrer-Caja, E., Hamagami, F., &
Woodcock, R. W. (2002) Comparative longi¬
tudinal structural analyses of the growth and
decline of multiple intellectual abilities over
the life span. Developmental Psychology, 38 (1),
115—142.
McDaniel, M. A. (2005) Big-brained people are
smarter: A meta-analysis of the relationship
between in vivo brain volume and intelli¬
gence. Intelligence, 33 (4), 337-346.
McDaniel, M. A., Moregson, F. P, Finnegan,
E. B., Campion, M. A., & Braverman, E. P.
(2001) Use of situational judgment tests to pre¬
dict job performance: A clarification of the lit¬
erature. Journal of Applied Psychology, 86 (4),
730-74°.
McGrew, K. S. (2005) The Cattell-Horn-Carroll
theory of cognitive abilities. In D. P.
Flanagan & P. L. Harrison, Contemporary
REFERENCES
473
intellectual assessment: Theories, tests, and
issues. New York: Guilford, pp. 136-181.
McGrew, K. S., Werder, J. K., & Woodcock,
R. W. (1991) Woodcock- Johnson technical man¬
ual. Allen, TX: DLM Teaching Resources.
McGrew, K. S., & Woodcock, R. W. (2001) Tech¬
nical manual: Woodcock Johnson III. Itasca, IL:
Riverside Press.
McGue, M. (1999) Behavior genetic models of
alcoholism and drinking. In K. E. Leonard &
H. T. Blaine (Eds.], Psychological theories of
drinking and alcoholism (2nd ed.). New York.
Guilford Press, pp. 372-421.
McGue, M., Bouchard, T. J., Jr., Iacono, W. G.,
& Lykken, D. T. (1993] Behavioral genetics of
cognitive ability: A life-span perspective. In
R. Plomin & G. E. McClearn (Eds.], Nature,
nurture, and psychology. Washington, DC:
American Psychological Association, pp. 59-
76.
McHenry, J. J., Hough, L. M., Toquam, J. L.,
Hanson, M. A., & Ashworth, S. (1990] Project
A validity results: The relationship between
predictor and criterion domains. Personnel
Psychology, 43 (2], 335-354.
McKay, P. F., & McDaniel, M. A. (2006] A reex¬
amination of black- white mean differences in
work performance: More data, more modera¬
tors. Journal of Applied Psychology, 91 (3], 538-
544-
Mekel-Bobrov, N., Gilbert, S. L., Evans, P. D.,
Vallender, E. J., Anderson, J. R., Hudson,
R. R., Tishkoff, S. A., & Lahn, B. T. (2005]
Ongoing adaptive evolution of ASPM, a brain
size determinant in Homo sapiens. Science, 309
(9 September], 1720-1722.
Mekel- Bobrov, N., Posthuma, D., Gilbert, S. L.,
Lind, P., Gosso, M. F., Luciano, M., Harris,
S. E., Bates, T. C., Polderman, T. J. C., Whal-
ley, L. J., Fox, H., Starr, J. M., Evans, P. D.,
Montgomery, G. W., Fernandes, C., Heutnik,
P., Martin, N. G., Boomsma, D. L, Deary, I.
J., Wright, M. J., de Geuss, E. J. C., & Lahn,
B. T. (2007] The ongoing adaptive evolution
of ASPM and Microcephalin is not explained
by increased intelligence. Human Molecular
Genetics, 16 (6], 600-608.
Mervis, C. B. (203] Williams syndrome: 15 years
of psychological research. Developmental Neu¬
ropsychology, 23 (1-2], 1-12.
Messick, S., & Jungeblut, A. (1981] Time and
method in coaching for the SAT. Psycholog¬
ical Bulletin, 89 (2], 191-216.
Meyer, D. E., & Kieras, D. E. (1997] A
computational theory of executive cognition
processes and multi-task performance. Part 1:
Basic mechanisms. Psychological Review, 104
(0, 3-65-
Milne, A. A. (1926] Winnie-the-Pooh. London:
Methuen.
Milner, B. (2005] The medial temporal-lobe
amnesic syndrome. Psychiatric Clinics of North
America, 28 (3], 599-611.
Milner, B., Corkin, S., & Teuber, H. L. (1968)
Further analysis of the hippocampal amnesic
syndrome: 14-year follow-up study of H. M.
Neuropsychologia, 6, 215-234.
Minstrell, J. (2001] The role of the teacher in mak¬
ing sense of classroom experience and effect¬
ing better learning. In S. M. Carver & D. Klahr
(Eds.], Cognition and instruction: 25 years of
progress. Mahwah, NJ: Erlbaum, pp. 121-150.
Moeser, S. D. (1988] Cognitive mapping in a com¬
plex building. Environment and Behavior, 20
(1], 21-49.
Moffat, S. D., Kennedy, K. M., Rodrigue, K. M.,
& Raz, N. (2007] Extrahippocampal contribu¬
tions to age differences in human spatial nav¬
igation. Cerebral Cortex, 17 (6), 1274-1282.
Molenaar, D., Dolan, C. V., & Wicherts, J. M.
(2009] The power to detect sex differences in
IQ test scores using Multi-Group Covariance
and Means Structural Analysis. Intelligence, 37
(4) . 396-4°4-
Molfese, V,, DeLalla, L. F., & Bunce, D. (1997]
Prediction of the intelligence test scores of 3-
to-8-year old children by home environment,
socioeconomic status, and biomedical risks.
Merrid-P aimer Quarterly, 41 (2), 219-234.
Mortensen, L. H., Sorensen, T. I. A., & Gron-
baek, M. (2005) Intelligence in relation to
later beverage preference and alcohol intake.
Addiction, 100 (10), 1445—1452.
Moselhy, H., Georgiou, G., & Kahn, A. (2001]
Frontal lobe changes in alcoholism: A review
of the literature. Alcohol and Alcoholism, 36
(5) - 357-?68-
Muller, C. B., et al. (72 additional signatures]
(2005] Gender difference and performance in
science. Science, 307 (18 February], 1043.
Murray, C. A. (1998] Income inequality and
IQ. Washington, DC: American Enterprise
Institute.
Murray, C. A. (2003] Human accomplishment.
New York: Harper.
Murray, C. A. (2005] The inequality taboo. Com¬
mentary, 120 (2], 13-22.
Murray, C. (2007] The magnitude and compo¬
nents of change in the black-white IQ differ¬
ence from 1920 to 1991: A birth cohort analysis
474
REFERENCES
of the Woodcock-Johnson standardizations.
Intelligence, 35, 305-318.
Murray, C. A (2008) Real education: Four sim¬
ple truths for bringing America's schools back to
reality. New York: Random House.
Myerson, J., Hale, S., Wagstaff, D., Poon, L. W.,
& Smith, G. A. [1990) The information-loss
model: A mathematical theory of age-related
cognitive slowing. Psychological Review, 97,
475-487.
Naglieri, J. A. (2005] The Cognitive Assessment
System. In D. P. Flanagan & L. Harrison
(Eds.), Contemporary intellectual assessment:
Theories, tests, and issues. New York: Guilford
Press, pp. 441-460.
Nathawat, S. S., & Puri, P. (1995) A compara¬
tive study of MZ and DZ twins on Level I
and Level II mental abilities and personality.
Journal of the Indian Academy of Applied Psy¬
chology, 21, 87-92.
Needleman, Herbert L., Gunnoe, C., Leviton,
A., Reed, R., Peresie, H., Maher, C., & Barret,
P. (1979) Deficits in psychologic and classroom
performance of children with elevated dentine
lead levels. New England Journal of Medicine,
300 (13), 689-695.
Neisser, U., Boodoo, G., Bouchard, T. J., Jr.,
Boykin, A. W., Brody, N., Ceci, S. J., Halpern,
D. F., Loehlin, J. C., Perloff, R., Sternberg,
R. J., & Urbina, S. (1996) Intelligence: Knowns
and unknowns. American Psychologist, 51 (2),
77-101.
Neisser, U. (Ed.) (1998) The rising curve: Longterm
gains in IQ and related measures. Washington,
DC: American Psychological Association.
Nettlebeck, T. (1987) Inspection time and intelli¬
gence. In P. A. Vernon (Ed.), Speed of informa¬
tion processing and intelligence. Norwood, NJ:
Ablex, pp. 295-346.
Nettlebeck, T. (2001) Correlations between
inspection time and psychometric ability: A
personal interpretation. Intelligence, 29, 459-
474-
Neubauer, A. (2000) Physiological approaches to
human intelligence. Psychologische Beitrage, 42
(2), 161-173.
Neubauer, A. C., Fink, A., & Schrausser, D. G.
(2001) Intelligence and neural efficiency: The
influence of task content and sex on brain-IQ
relationship. Intelligence, 30 (6), 515-536.
Neubauer, A. C., Grabner, R. H., Fink, A., &
Neuper, C. (2005) Intelligence and neural effi¬
ciency: Further evidence of the influence of
task content and sex on the brain-IQ relation¬
ship. Cognitive Brain Research, 25 (1), 217-225.
Newcombe, N. S. (2007) Taking science seri¬
ously: Straight thinking about spatial sex
differences. In S. J. Ceci & W. M. Williams
(Eds.), Why aren't there more women in science:
Top researchers debate the evidence. Washing¬
ton, DC: American Psychological Association,
pp. 69-77.
Newcombe, R., Milne, B. J., Caspi, A., Poulton,
R., & Moffitt, T. E. (2007) Birth weight pre¬
dicts IQ: Fact or artifact? Twin Research and
Human Genetics, 10 (4), 581-586.
Newell, A. (1990) Unified theories of cognition.
Cambridge, MA: Harvard University Press.
Nguyen, H.-H. D., & Ryan, A. M. (2008) Does
stereotype threat affect test performance of
minorities and women? A meta-analysis of
experimental evidence. Journal of Applied
Psychology, 93 (6), 1314-1334.
Nickerson, R. S. (2010) Mathematical reasoning:
Patterns, problems, conjectures, and proof New
York: Psychology Press.
Nisbett, R. E. (2009) Intelligence and how to get it.
New York: Norton.
Nyborg, H. (2005) Sex-related differences in gen¬
eral intelligence, g, brain size and social status.
Personality and Individual Differences, 39 (3),
497-509.
Nyborg, H., & Jensen, A. R. (2000) Black- white
differences on various psychometric tests:
Spearman's hypothesis tested on American
Armed Services veterans. Personality and Indi¬
vidual Differences, 28 (3), 593-599.
Nyborg, H., & Jensen, A. R. (2001) Occupation
and income related to psychometric g. Intelli¬
gence, 29 (1), 45-55.
Oberauer, K. (2003) The multiple faces of work¬
ing memory: Storage, processing, and co¬
ordination. Intelligence, 31 (2), 167-193.
Ogbu, J. (2003) Black American students in an
affluent suburb: A study of academic disengage¬
ment. Mahwah, NJ: Erlbaum.
O' Grady, K. E. (1983) A confirmatory maximum
likelihood factor analysis of the WAIS-R. Jour¬
nal of Consulting and Clinical Psychology, 51
(6), 826-831.
Ohnishi, T., Matsuda, H., Hirakata, M., &
Ugawa, Y. (2006) Navigation ability depen¬
dent neural activation in the human brain: An
fMRI study. Neuroscience Research, 55 (4), 361—
369.
O’Toole, B. I. (1990) Intelligence and behaviour
and motor vehicle accident mortality. Accident
Analysis and Prevention, 22 (3), 211-221.
Pal, S., Shyam, R., & Singh, R. (1997) Genetic
analysis of general intelligence ‘g’: A twin
REFERENCES
475
study. Personality and Individual Differences,
^ (5). 779~78°-
Palmer, J., MacLeod, C. M., Hunt, E., & David¬
son, J. (1985) Information processing correlates
of reading. Journal of Memory and Language,
24, 59-88.
Park, G., Lubinski, D., & Benbow, C. P. (2006)
Ability differences among people who have
commensurate degrees matter for scientific
productivity. Psychological Science, 19 [10),
957-961.
Park, G., Lubinski, D., & Benbow, C. P. (2007}
Contrasting intellectual patterns predict cre¬
ativity in the arts and sciences: Tracking intel¬
lectually precocious youth over 25 years. Psy¬
chological Science, 10 (11) , 948-952.
Parker, E. S., Parker, D. A., & Harford, T. C.
(1991) Specifying the relationship between
alcohol use and cognitive loss: The effects of
frequency of consumption and psychological
distress. Journal of Studies on Alcohol, 52 (4),
366-373.
Pennock-Roman, M. (1994) College major and
gender differences in the prediction of college
grades. College Board Report 94-2. New York:
College Entrance Examination Board.
Pentland, A. (2010) To signal is human. American
Scientist, 98 (May-June), 204-211.
Peter, L. J. (1969) The Peter Principle. New York:
Morrow.
Petrill, S. A. (2002) The case for general intel¬
ligence: A behavior genetics perspective. In
R. J. Sternberg & E. L. Grigorenko (Eds.}, The
general factor of intelligence: How general is it?
Mahwah, NJ: Erlbaum, pp. 281-298.
Petrill, S. A., Lipton, P. A., Hewitt, J. K., Plomin,
R., Cherny, S. S., Corley, R., & DeFries, J. C.
(2004} Genetic and environmental contribu¬
tions to general cognitive ability through the
first 16 years of life. Developmental Psychology,
40 (5), 805-812.
Phillips, M., Brooks-Gunn, J., Duncan, G. J.,
Klebanov, P., & Crane, J. (1998} Family back¬
ground, parenting practices, and the Black-
White test score gap. In C. Jenks & M. Phillips
(Eds.), The black-white test score gap. Washing¬
ton, DC: Brookings Institute, pp. 103-145.
Phillips, M., Crouse, J., & Ralph, J. (1998) Does
the black- white test score gap widen after chil¬
dren enter school? In C. Jenks & M. Phillips
(Eds.), The black-white test score gap. Wash¬
ington, DC: Brookings Institute, pp. 229-
272.
Piazza, M., & Dehaene, S. (2004) From num¬
ber neurons to mental arithmetic: The cogni¬
tive neuroscience of number sense. In M. S.
Gazzaniga (Ed.), The cognitive neurosciences
(3rd edition). Cambridge, MA: MIT Press,
pp. 865-875.
Pind, J., Gunnarsdottir, E. K., & Johannesson,
H. S. (2003) Raven’s standard progressive
matrices: New school age norms and a study
of the test’s validity. Personality and Individual
Differences, 34, 375-386.
Plomin, R., DeFries, J. C., McClearn, G. E., &
McGuffin, P. (2008) Behamoral Genetics (5th
edition). New York: Worth Publishers.
Plomin, R., Fulker, D. W., Corley, R., & DeFries,
J. C. (1997) Nature, nurture, and cognitive
development from 1 to 16 years: A parent¬
offspring adoption study. Psychological Science,
8 (6), 442-447.
Plomin, R., Kennedy, J. K. J., & Craig, I. W.
(2006) The quest for quantitative trait loci
associated with intelligence. Intelligence, 34
(6); 513“526*
Plucker, J. A. (1999) Is the proof in the pudding?
Reanalysis of Torrance’s (1958 to present) lon¬
gitudinal data. Creativity Research Journal, 12
(2), 103-114.
Polderman, T. J. C., Posthuma, D., De Son-
neville, L. M. J., Stins, J. F., Verhulst, F. C., &
Boomsma, D. I. (2007) Genetic analysis of the
stability of executive functioning during child¬
hood. Biological Psychology, 76 (1-2), n-20.
Polderman, T. J. C., Stins, J. F., Posthuma, D.,
Gasso, M. F., Verhulst, F. C., & Boomsma,
D. I. (2006) The phenotypic and genotypic
relation between working memory speed and
capacity. Intelligence, 34 (6), 549-560.
Pollitt, E., Gorman, K. S., Patrice, L., Martorell,
R., et al. (1993) Early supplementary feed¬
ing and cognition: Effects over two decades.
Monographs of the Society for Research in Child
Development, 58 (7), v-99.
Poltrock, S. E. (1977) Models of temporal order
discrimination. Unpublished Ph.D. disserta¬
tion, University of Washington.
Poltrock, S. E., & Brown, P. (1984) Individual dif¬
ferences in visual imagery and spatial ability.
Intelligence, 8 (2), 93-138.
Poretus, S. D., Dewey, D. M., & Bernreuter,
R. G. (1930) Race and social differences in
performance tests. Genetic Psychology Mono¬
graphs, 8, 95-208.
Posner, M. I., Boies, S. E., Eichelman, W. H.,
& Taylor, R. L. (1969) Retention of visual
and name codes of single letters. Journal of
Experimental Psychological Monographs, 79, 1-
16.
476
REFERENCES
Posner, M. I., & Rothbart, M. K. (2007) Educating
the human brain. Washington, DC: American
Psychological Association.
Posner, M. I., Sheese, B. E., Odludas, Y., & Tang,
Y. Y. (2006) Analyzing and shaping human
attentional networks. Neural Networks, 19 (9),
1422-1429.
Posthuma, D., de Geus, E. J. C., & Boomsma,
D. I. (2001) Perceptual speed and IQ are associ¬
ated through common genetic factors. Behav¬
ior Genetics, 31 (6), 593-602.
Posthuma, D., de Geus, E. J. C., & Boomsma,
D. I. (2004] Genetic contributions to anatomi¬
cal, behavioral, and neurophysiological indices
of cognition. In R. Plomin, J. C. DeFries, I. W.
Craig, & P. McGuffin (Eds.), Behavioral genet¬
ics in the postgenomic era. Washington, DC:
American Psychological Association, pp. 141-
161.
Posthuma, D., Luciano, M., de Geus, E. J. C.,
Wright, M. J., Slagboom, P. E., Montgomery,
G. W., Boomsma, D. I., & Martin, N. G. (2005)
A genomewide scan for intelligence identifies
quantitative trait loci on 2q and 6p. American
Journal of Human Genetics, 77, 318-326.
Prat, C. S., Keller, T. A., & Just, M. A. (2007)
Individual differences in sentence compre¬
hension: A functional magnetic resonance
imaging investigation of syntactic and lexical
processing demands. Journal of Cognitive Neu¬
roscience, 19 (12), 1950-1963.
Preston, S. H. (1998) Differential fertility by IQ
and the IQ distribution of a population. In U.
Neisser (Ed.), The rising curve: Long-term gains
in IQ and related measures. Washington, DC:
American Psychological Association, pp. 317-
338.
Pulakos, E. D., Schmitt, N. A., & Chan, D. (1996)
Models of job performance: An examination
of ratee race, ratee gender, and rater level
effects. Human Performance, 9 (2), 103-119.
Puts, D. A., McDaniel, M. A., Jordan, C. L., &
Breedlove, S. M. (2008) Spatial ability and pre¬
natal androgens: Meta-analyses of congenital
adrenal hyperplasia and digit ratio (2D:qD).
Archives of sexual behavior, 37 (1), 100-
111.
Raftery, A. E. (1995) Bayesian model selection in
sociological research. Sociological Methodology,
25, 111-163.
Ramist, L., Lewis, C., & McCamley-Jenkins, L.
(1994) Student group differences in predict¬
ing college grades: Sex, language, and ethnic
groups. College Board Research Report 93-
1. New York: College Entrance Examination
Board.
Rauscher, F. H., Shaw, G. L., & Ky, K. N. (1993)
Music and spatial task performance. Nature,
365 (14 October), 365.
Raven, J. (1989) The Raven Progressive Matri¬
ces: A review of national norming studies and
ethnic and socioeconomic variation within the
United States. Journal of Educational Measure¬
ment, 26 (1), 1-16.
Raven, J. (2000) The Raven's Progressive Matri¬
ces: Change and stability over culture and
time. Cognitive Psychology, 41 (1), 1-48.
Raven, J. (2008a) Change and stability in RPM
scores over culture and time: The story at the
turn of the century. In J. Raven & J. Raven
(Eds.), Uses and abuses of intelligence: Spear
man and Raven’s Quest for Non- Arbitrary Met¬
rics. New York: Royal Fireworks Press, pp. 213-
258.
Raven, J. (2008b) The Raven Progressive Matri¬
ces Tests: Their theoretical basis and measure¬
ment model. In J. Raven & J. Raven (Eds.),
Uses and abuses of intelligence: Spearman and
Raven’s Quest for Non- Arbitrary Metrics. New
York: Royal Fireworks Press, pp. 14-68.
Raven, J. (2008c) Intelligence, engineered abil¬
ity, and the destruction of life on Earth. In
J. Raven & J. Raven (Eds.), Uses and abuses
of intelligence: Spearman and Raven’s Quest for
Non- Arbitrary Metrics. New York: Royal Fire¬
works Press, pp. 431-471.
Raven, J. (2oo8d) Psychometrics, cognitive abil¬
ity, and occupational performance. In J. Raven
& J. Raven (Eds.), Uses and abuses of intelli¬
gence: Spearman and Raven's Quest for Non -
Arbitrary Metrics. New York: Royal Fireworks
Press, pp. 472-518.
Raven, J., & Raven, J. (Eds.), (2008) Uses and
abuses of intelligence: Studies advancing Spear¬
man and Raven’s quest for non-arbitrary met¬
rics. New York: Royal Fireworks Press.
Raven, J. C. (1938) Progressive Matrices: A per¬
ceptual test of intelligence. London: H. K.
Lewis.
Raz, N., Lindenberger, U., Ghisletta, P.,
Rodrigue, K. M., Kennedy, K. M.# & Acker,
J. D. (2008) Neuroanatomical correlates of
fluid intelligence in healthy adults and persons
with vascular risk factors. Cerebral Cortex, 18
(3)/ 7l8"726-
Reader, S. M., & Laland, K. N. (2002) Social intel¬
ligence, innovation, and enhanced brain size in
primates. Proceedings of the National Academy
of Sciences, 99, 4436—4441.
Ree, M. J., & Earles, J. A. (1991) The stability
of g across different methods of estimation.
Intelligence, 15 (3), 271-278.
REFERENCES
477
Reich, R. (1991) The work of nations; Preparing
ourselves for 2 1st century capitalism . New Y ork:
Knopf.
Reichle, E. D., Carpenter, P. A., & Just, M. A.
(2000] The neural basis of strategy and skill
in sentence-picture verification. Cognitive Psy¬
chology, 40 (3), 261-295.
Reynolds, C., Baker, L. A., & Pedersen, N. L.
(2000) Multivariate models of mixed assort¬
ment: Phenotypic assortment and social
homogamy for education and fluid ability.
Behavior Genetics, 30 (6), 455-476.
Reynolds, C. A., Finkel, D., McArdle, J. J., Gatz,
M., Berg, S., & Pedersen, N. L. (2005) Quan¬
titative genetic analysis of latent growth curve
models of cognitive abilities in adulthood.
Developmental Psychology, 41 (1), 1-16.
Rhode, T. E., & Thompson, L. A. (2007) Pre¬
dicting academic achievement with cognitive
ability. Intelligence, 35 (1), 83-92.
Rindermann, H. (2007} The g factor of inter¬
national cognitive ability comparisons: The
homogeneity of results in PISA, TIMSS,
PIRLS, and IQ tests across nations. European
Journal of Personality, 21 (5), 667—706.
Rindermann, H. (2008a) Relevance of education
and intelligence at the national level for the
economic welfare of people. Intelligence, 36
(2), 127-142.
Rindermann, H. (2008b) Relevance of educa¬
tion and intelligence for the political develop¬
ment of nations: Democracy, the rule of law,
and political liberty. Intelligence, 36 (4), 306-
322.
Rindermann, H., & Ceci, S. J. (2009) Educational
policy and country outcomes in international
cognitive competence studies. Perspectives on
Psychological Science, 4 (6), 551-577.
Rips, L. J. (1994) The psychology of proof : Deductive
reasoning in human thinking. Cambridge, MA:
MIT Press.
Robb, G. P., Bernardoni, L. C., & Johnson, R. W.
(1972) Assessment of indimdual mental ability.
Scranton, PA: Intext Educational Publishers.
Roberts, R. D., Goff, G. N., Anjoul, F., Kyllo-
nen, P. C., Pallier, G., & Stankov, L. (2000)
The Armed Services Vocational Aptitude Bat¬
tery (ASVAB): Little more than acculturated
learning (Gc)!? Learning and Individual Differ¬
ences, 12 (1), 81-103.
Robinson, N. M. (1996) Acceleration as an option
for the highly gifted adolescent. In C. P. Ben-
bow & D. J. Lubinski (Eds.), Intellectual tal¬
ent: Psychometric and social issues. Baltimore,
MD: Johns Hopkins University Press, pp. 169-
178.
Rodgers, J. L., Cleveland, H. H., van den Ourd,
E., & Rowe, D. C. (2000) Resolving the debate
over birth order, family size, and intelligence.
American Psychologist, 55 (6), 599-612.
Rosen, A. C., Prull, M. W., Gabrieli, J. D.
E. , Stoub, T., O’Hara, R., Friedman, L.,
Yesavage, J. A., & deToledo-Morrell, L. (2003)
Differential associations between entorhinal
and hippocampal volumes in older adults.
Behavioral Neuroscience, 117 (6), 1150-1160.
Rosenberg, N. A., Pritchard, J. K., Weber, J. L.,
Cann, H. M., Field, K. K., Zhivotovsky, L. A.,
& Feldman, M. A. (2002) Genetic structure of
human populations. Science, 298 (20 Decem¬
ber), 2381-2385.
Ross, J., Zinn, A., & McCauley, E. (2000) Neu-
rodevelopmental and psychosocial aspects of
Turner Syndrome. Mental Retardation and
Developmental Disabilities Research Reviews, 6,
135_141-
Roth, P. L., Bevier, C. A., Bobko, P., Switzer,
F. S. Ill, & Tyler, P. (2001) Ethnic group differ¬
ences in cognitive ability in employment and
educational settings: A meta-analysis. Person¬
nel Psychology, 54 (2), 297-330.
Roth, P. L., Huffcutt, A. I., & Bobko, P. (2003)
Ethnic group differences in measures of job
performance: A new meta-analysis. Journal of
Applied Psychology, 88 (4), 694-706.
Rowe, D. C. (2004) Assessing genotype-
environment interactions and correlations in
the postgenomic era. In R. Plomin, J. C.
DeFries, I. W. Craig, & P. McGuffm (Eds.),
Behavioral genetics in the postgenomic era .
Washington, DC: American Psychological
Association, pp. 71-86.
Rushton, J. P. (1992) Cranial capacity related to
sex, rank, and race in a stratified sample of
6325 U. S. Military personnel. Intelligence, 16,
401-413.
Rushton, J. P. (1994) Sex and race differences
in cranial capacity from international labour
office data. Intelligence, 19 (3), 281-294.
Rushton, J. P. (2008) Testing the genetic hypoth¬
esis of group mean IQ differences in South
Africa: Racial admixture and cross-situational
consistency. Personality and Individual Differ¬
ences, 44 (3), 768-776.
Rushton, J. P., & Ankney, C. D. (1996) Brain size
and cognitive ability: Correlations with age,
sex, social class, and race. Psychonomic Bulletin
and Reinew, 3 (1), 21-36.
Rushton, J. P., Cvorovic, J., & Bons, T. A. (2007)
General mental ability in South Asians: Data
from three Roma (Gypsy) communities in
Serbia. Intelligence, 35 (1), 1-12.
478
REFERENCES
Rushton, J. P.; & Jensen, A. R. (2005) Thirty years
of research on race differences in cognitive
ability. Psychology, Public Policy , and Law, 11
00, 235-294.
Rushton, J. P., & Osborne, R. T. (1995) Genetic
and environmental contributions to cranial
capacity in Black and White adolescents. Intel
ligence, 20 (1), 1-13.
Rushton, J. P., & Skuy, M. (2000) Performance on
Raven’s Matrices by African and White uni¬
versity students in South Africa. Intelligence ,
28 (4], 251-265.
Rushton, J. P., Skuy, M., & Fridjhon, P. (2003)
Performance on Raven's Advanced Progres¬
sive Matrices by African, East Indian, and
White engineering students in South Africa.
Intelligence, 31 (2), 123-137.
Rushton, J. P., Vernon, P. A., & Bons, T. A. (2007)
No evidence that polymorphisms of brain
regulator genes Microcephalin and ASPM are
associated with general mental ability, head
circumference, or altruism. Evolutionary Biol¬
ogy , 3 00, 157-160.
Sackett, P. R., Bomeman, M. J., & Connelley,
B. S. (2008) High-stakes testing in higher edu¬
cation and employment. American Psycholo¬
gist, 6 (4), 215-227.
Sackett, P. R., Hardison, C. M., & Cullen,
M. J. (2004] On interpreting stereotype threat
as accounting for African American-White
differences on cognitive tests. American Psy¬
chologist, 59 (1), 7-13.
Sackett, P. R., Kuncel, N. R., Arenson, J. J.,
Cooper, S. R., & Waters, S. D. (2009] Does
socioeconomic status explain the relationship
between admissions tests and post-secondary
academic performance? Psychological Bulletin,
135 to, 1-22.
Saklofske, D. H., Yang, Z., Zhu, J., & Austin, E. J.
(2008] Spearman's law of diminishing returns
in normative samples for the WISC-IV and
WAIS-III. Journal of Individual Differences, 29
(1), 57-69.
Salovey, P., & Grewal, D. (2005) The science of
emotional intelligence. Current Directions in
Psychological Science, 14 (6), 281-285.
Salovey, P., & Mayer, J. D. (1990) Emotional
intelligence. Imagination , Cognition and Per¬
sonality, 9 (3), 185-211.
Salthouse, T. A. (1996] The processing-speed the¬
ory of adult age differences in cognition. Psy¬
chological Review, 103 (3), 403-428.
Salthouse, T. A. (2005) Relations between cogni¬
tive abilities and measures of executive func¬
tioning. Neuropsychology, 19 (4], 532-545.
Salthouse, T. A., Atkinson, T. M., & Berish, D.
E. (2003) Executive functioning as a poten¬
tial mediator of age-related cognitive decline
in normal adults. Journal of Experimental Psy¬
chology: General, 132 (4), 566-594.
Sampson, P. D., Kerr, B., Olson, H. C.,
Streissguth, A. P., Hunt, E. B., & Barr, H. M.
(1997) The effects of prenatal alcohol exposure
on adolescent cognitive processing: A speed-
accuracy tradeoff. Intelligence, 24 (2), 329-
353*
Santy, P. A. (1994} Choosing the right stuff: The
psychological selection of astronauts and cosmo¬
nauts. Westport, CT: Praeger.
Sarason, I. G. (1980) Test anxiety: Theory, research,
and applications. Mahwah, NJ: Erlbaum.
Scarr, S., & Weinberg, R. A. (1976) IQ perfor¬
mance of black children adopted by white
families. American Psychologist, 31, 726-739.
Scarr, S., & Weinberg, R. A. [1977] Intellectual
similarities within families of both adopted
and biological children. Intelligence, 1 (2), 170-
191.
Scarr, S., & Weinberg, R. A. (1983) The Min¬
nesota Adoption Studies: Genetic differences
and malleability. Child Development, 54 (2],
260-267.
Schaie, K. W. (1996) Intellectual development in
adulthood: The Seattle longitudinal study. Cam¬
bridge: Cambridge University Press.
Schaie, K. W. (2005) Developmental influences on
adult intelligence. Oxford: Oxford University
Press.
Schaie, K. W., & Strother, C. R. (1968] A cross-
sequential study of age changes in cognitive
behavior. Psychological Bulletin, 70 [6], 671-
680.
Schaler, J. A. (Ed.) (2006) Howard Gardner
underfire: The rebel psychologist faces his critics.
Chicago: Open Court.
Schiff, M., & Lewontin, R. (1986) Education and
class. Oxford: Clarendon Press.
Schmidt, F. L., & Hunter, J. E. (1998) The validity
and utility of selection methods in personnel
psychology: Practical and theoretical implica¬
tions of 85 years of research findings. Psycho¬
logical Bulletin, 124 (2), 262-274.
Schnakenberg, R. (2004) Distory: A treasury of his¬
torical insults. New York: St. Martin's Press.
Schneider, W., & Shiffrin, R. M. (1977) Con¬
trolled and automatic processing: I. Detection,
search, and attention. Psychological Review, 84
(1), 1-66.
Schooler, C. (1998) Environmental complexity
and the Flynn effect. In U. Neisser (Ed.), The
REFERENCES
479
rising curve: Long-term gains in IQ and related
measures. Washington, DC: American Psy¬
chological Association, pp. 67-79.
Schottenbauer, M. A., Momenan, R. M., Kerick,
M. H., & Hommer, D. W. (2007) Rela¬
tionships among aging, IQ, and intracranial
volume in alcoholics and control subjects.
Neuropsychology, 3, 337-345.
Schrank, F. A. (2005) The Woodcock-Johnson III
tests of cognitive abilities. In D. P. Flanagan
& P. L. Harrison (Eds.), Contemporary intellec¬
tual assessment: Theories, tests, and issues. New
York: Guilford, pp. 371-401.
Scullin, M. H., Peters, E., Williams, W. M., &
Ceci, S. J. (2000) The role of IQ and edu¬
cation in predicting later labor market out¬
comes. Psychology, Public Policy, and the Law,
^ 6 (1), 63-89.
Segal, N. L. (2000) Virtual twins: New findings
on within-family environmental influences on
intelligence. Journal of Educational Psychology,
92 (3)/ 442-44§-
Segal, N. L., & Hershberger, S. L. (2005) Vir¬
tual twins and intelligence: Updated and new
analyses of within-family environmental influ¬
ences. Personality and Individual Differences,
39 (6), 1061-1073.
Semendeferi, K., Lu, A., Schenker, N., & Dama-
sio, H. (2002) Humans and great apes share
a large frontal cortex. Nature Neuroscience, 5
(3), 272-276.
Shallice, T. (2004) The fractionation of supervi¬
sory control. In M. S. Gazzaniga (Ed.), The
cognitive neurosciences (3rd ed.). Cambridge,
MA: MIT Press, pp. 943-956.
Shapiro, F. R. (2006) The Yale book of quotations:
New Haven, CT: Yale University Press.
Shaw, P., Greenstein, D., Lerch, J. Clasen, L.,
Lenroot, R., Gogtay, N., Evans, A., Rapoport,
J., & Giedd, J. (2006) Intellectual ability and
cortical development in children and adults.
Nature, 440 (30 March), 676-679.
Shepard, L. D., & Vernon, P. A., (2008) Intelli¬
gence and speed of information-processing: A
review of 50 years of research. Personality and
Individual Differences, 44 (3), 535-551.
Shields, J., Konold, T. R., & Glutting, J. J. (2004)
Validity of the Wide Range Intelligence Test:
Differential effects across race/ethnicity, gen¬
der, and education level. Journal of Psychoed-
ucational Assessment, 22, 287-303.
Shih, M., Pittinsky, T. L., & Ambady, N. (1999)
Stereotype susceptibility: Identity salience
and shifts in quantitative performance. Psy¬
chological Science, 10 (1), 80-83.
Shoda, Y., Mischel, W., & Peake, P. K.
(1990) Predicting adolescent cognitive and
self-regulatory competencies from pre-school
delay of gratification: Identifying diagnostic
conditions. Developmental Psychology, 26 (6),
978-986.
Shute, V. J., Gawlick-Grendell, L. A., Young,
R. K., & Burnham, C. (1996) An experien¬
tial system for learning probability: Stat Lady
description and evaluation. Instructional Sci¬
ence, 24 (1), 25-46.
Sigman, M., & Whaley, S. E. (1998) The role of
nutrition in the development of intelligence.
In U. Neisser (Ed.), The rising curve: Long term
gains in IQ and related measures. Washing¬
ton, DC: American Psychological Association,
pp. 155-182.
Simonton, D. K. (1984) Genius, creativity,
and leadership : Historiometric inquiries. Cam¬
bridge, MA: Harvard University Press.
Simonton, D. K. (2001) Predicting presidential
performance in the United States: Equation
replication and recent survey results. Journal
of Social Psychology, 141 (3), 293-308.
Simonton, D. K. (2006) Presidential IQ, open¬
ness, intellectual brilliance, and leadership:
Estimates and correlations for 42 U. S. chief
executives. Political Psychology , 27 (4), 511-
526.
Skarbrevik, K. J. (2002) Gender differences
among students found eligible for special edu¬
cation. European Journal of Special Needs Edu¬
cation, 17 (2), 97-107.
Skodak, M., & Skeels, H. M. (1949) A final follow¬
up study of one hundred adopted children.
Journal of General Psychology, 75, 85-125.
Smedley, A., & Smedley, B. D. (2005) Race as
biology is fiction, racism as a social problem
is real: Anthroplogical and historical perspec¬
tives on the social construction of race. Amer¬
ican Psychologist, 60 (i), 16-26.
Smith, D. I., & Kirkham, R. W. (1982) Relation¬
ship between intelligence and driving record.
Accident Analysis and Prevention, 14 (6), 439-
442.
Smith, E. E., & Jonides, J. (1999) Storage and
executive processes in the frontal lobes. Sci¬
ence, 283 (5408), 1657-1661.
Snow, R. E. (1982) Education and intelligence.
In R. J. Sternberg (Ed.), Handbook of human
intelligence. New York: Cambridge University
Press, pp. 493-585.
Snow, R. E. (1996) Aptitude development and
education. Psychology, Public Policy, and Law,
2 (3/4). 536-560.
480
REFERENCES
Snow, W. G., & Weinstock, J. (1990) Sex dif¬
ferences among non-brain-damaged adults on
the Wechsler Adult Intelligence Scales: A
review of the literature. Journal of Clinical and
Experimental Neuropsychology , 126 (6), 873—
886.
Song, C., Qu, Z., Blumm, N., & Barabasi,
A.-L. [2010) Limits of predictability in human
mobility. Science, 327 (5698), 1018-1021.
Spearman, C. (1904) General intelligence, objec¬
tively determined and measured. American
Journal of Psychology, 15, 201-293.
Spearman, C. [1923) The nature of {iintelli-
gence” and the principles of cognition. London:
Methuen.
Spearman, C. (1927) The abilities of man. London:
MacMillan.
Spelke, E. S. (2005] Sex differences in intrinsic
aptitude for mathematics and science? A crit¬
ical review. American Psychologist, 60 (9), 950-
958.
Stankov, L & Horn, J. L. (1980) Human abili¬
ties revealed through auditory tests. Journal of
Educational Psychology, 72 (1), 21-44.
Stanley, J. (1996) In the beginning: The study of
mathematically precocious youth (SMPY). In
C. P. Benbow 8l D. Lubinski (Eds.), Intellec¬
tual talent: Psychometric and social issues. Bal¬
timore, MD: Johns Hopkins University Press,
pp. 225-235.
Steele, C. M., & Aaronson, J. (1995) Stereotype
threat and the intellectual test performance of
African Americans. Journal of Personality and
Social Psychology, 69, 791-811.
Steele, C. M., & Aaronson, J. (1998) Stereo¬
type threat and the test performance of aca¬
demically successful African Americans. In
C. Jencks & P. Meredith (Eds.), The black-
white test score gap. Washington, DC: Brook¬
ings, pp. 401-427.
Stein, Z., Susser, M., Saenger, G., & Marolla,
F. (1972) Nutrition and mental performance.
Science, 178 (4062), 708-713.
Steinberg, R., with Brown, B. B., & Dornbusch,
S. M. (1996) Beyond the classroom: Why school
reform has failed and what parents need to do.
New York: Simon & Schuster.
Sternberg, R. J. (1977a) Intelligence, information
processing, and analogical reasoning: The com-
ponential analysis of human abilities. Hillsdale,
NJ: Erlbaum.
Sternberg, R. J. (1977b) Component processes in
analogical reasoning. Psychological Review, 84,
353"378-
Sternberg, R. J. (1981) Intelligence and entrench¬
ment. Journal of Educational Psychology , 73 (1),
1-16.
Sternberg, R. J. (1990) Metaphors of mind. Cam¬
bridge: Cambridge University Press.
Sternberg, R. J. (1996) Successful intelligence: How
practical and creative intelligence determines
success in life. New York: Simon & Schuster.
Sternberg, R. J. (2003a) Wisdom, intelligence, and
creativity synthesized. Cambridge: Cambridge
University Press.
Sternberg, R. J. (2003b) Issues in the theory
and measurement of successful intelligence:
A reply to Brody. Intelligence, 31 (4), 331-338.
Sternberg, R. J. (2004) Culture and intelligence.
American Psychologist, 59 (5), 325-338.
Sternberg, R. J. (2006) The Rainbow Project:
Enhancing the SAT through assessments of
analytical, practical, and creative skills. Intelli¬
gence, 34 (4), 312-350.
Sternberg, R. J. (2007) Finding students who are
wise, practical, and creative. The Chronicle of
Higher Education, July 6, 2007, 1311.
Sternberg, R. J., Ferrari, M., Clinkenbeard,
P. R., & Grigorenko, E. L. (1996) Identifi¬
cation, assessment, and instruction of gifted
children: A construct validation of a triarchic
model. Gifted Child Quarterly, 40, 129—137.
Sternberg, R. J., Forsythe, G. B., Hedlund, J.,
Horvath, J. A., Wagner, R. K., Williams,
W. M., Snook, S. A., & Grigorenko, E. L.
(2000) Practical intelligence in everyday life.
New York: Cambridge University Press.
Sternberg, R. J., & Grigorenko, E. L. (2007) The
difficulty of escaping preconceptions in writ¬
ing an article about the difficulty of escap¬
ing preconceptions: Commentary on Hunt
and Carlson (2007). Perspectives in Psycholog¬
ical Science, 2 (2), 221-223.
Sternberg, R. J., Grigorenko, E. L., Ferrari, M.,
& Clinkenbeard, P. (1999) A triarchic analy¬
sis of an aptitude treatment interaction. Euro¬
pean Journal of Psychological Assessment, 15 (1),
1-11.
Sternberg, R. J., Grigorenko, E. L., & Kidd, K. K.
(2005) Intelligence, race, and genetics. Ameri¬
can Psychologist, 60 (1), 46-59.
Sternberg, R. J., Grigorenko, E. L., & Zhang, L.-F.
(2008) Styles of learning and thinking matter
in instruction and assessment. Perspectives on
Psychological Science, 3 (6), 486-506.
Sternberg, R. J., Nokes, K., Geissler, P. W.,
Prince, R., Katcha, F., Bundy, D. A., & Grig¬
orenko, E. L. (2001) The relationship between
REFERENCES
481
academic and practical intelligence: A case
study in Kenya. Intelligence, 29, 401-418.
Sternberg, S. [1966) High speed scanning in
human memory. Science, 153, 652-654.
Sticht, T. G. (Ed.) (1975) Reading for working : A
functional literacy anthology: Alexandria, VA:
Human Resources Research Organization.
Sticht, T. G., Armstrong, W. B., Hickey, D. T., &
Cay lor, J. S. (1987) Cast-off youth: Policy and
training methods from the military experience.
New York: Praeger.
Stoolmiller, M. (1999) Implications of the
restricted range of family environments
for estimates of heritability and nonshared
environment in behavior-genetic adoption
studies. Psychological Bulletin, 125 (4), 392-
409.
Strand, S., Deary, I. J., & Smith, P. (2006) Sex
differences in cognitive ability tests: A UK
national picture. British Journal of Educational
Psychology, 76, 463-480.
S trickier, L. J., & Ward, W. C. (2004) Stereo¬
type threat, inquiring about test takers’ eth¬
nicity and gender, and standardized test per¬
formance. Journal of Applied Social Psychology,
34 (4)1 665-693.
Sue, S., & Okazaki, S. (1990) Asian- American
educational achievements: A phenomenon in
search of an explanation. American Psycholo¬
gist, 45 (8) (August), 913-920.
Sullivan, E. V., & Marsh, L. (2003) Hippocampal
volume deficits in alcoholic Korsakoff s syn¬
drome. Neurology, 61 (12), 1716-1719.
Sunderland, A., Harris, J. E., & Baddeley, A. D.
(1983) Do laboratory tests predict everyday
memory? A neuropsychological study. Jour¬
nal of Verbal Learning and Verbal Behavior, 22
(3)/ 341_357-
Suzuki, L., & Aaronson, J. (2005) The cultural
malleability of intelligence and its impact on
the racial/ethnic hierarchy. Psychology, Public
Policy, and the Law, 11 (2), 320-327.
Tang, H., Quertermous, T., Rodriguez, B.f
Kardia, S. L. R., Zhu, X., Brown, A.,
Pankow, J. S., Province, M. A., Hunt, S.
C.; Boerwinkle, E., Schork, N. J., & Risch,
N. J. (2005) Genetic structure, self-identified
race/ethnicity, and confounding in case-
control association studies. American Journal
of Human Genetics, 76, 268-275.
Tannenbaum, A. J. (1996) The IQ controversy
and the gifted. In C. P. Benbow & D. Lubin-
ski (Eds.), Intellectual talent . Baltimore: Johns
Hopkins University Press, pp. 44-82.
te Nijenhuis, J., De Jong, M.-J., Evers, A., &
Van Der Flier, H. (2004) Are cognitive dif¬
ferences between immigrants and majority
groups diminishing? European Journal of Per¬
sonality, 18, 405-434.
te Nijenhuis, J., van Vianen, A. E. M, & van der
Flier, H. (2007) Score gains on g-loaded tests:
No g. Intelligence, 35, 283-300.
Teasdale, T. W., & Owen, D. R. (1986) The influ¬
ence of paternal social class on intelligence and
educational level in male adoptees and non¬
adoptees. British Journal of Educational Psy¬
chology, 56 (1), 3-12.
Teasdale, T. W., & Owen, D. R. (2000) Forty-
year secular trends in cognitive abilities. Intel¬
ligence, 28 (2), 115-120.
Teasdale, T. W., & Owen, D. R. (2008) Secular
declines in cognitive test scores: A reversal of
the Flynn Effect. Intelligence, 36 (2), 121-126.
Templer, D., & Arikawa, H. (2006) Temperature,
skin color, per capita income, and IQ: An
international perspective. Intelligence, 34 (2),
121-127.
Terman, L. M., & Oden, M. H. (1959) The gifted
group at mid-life: Thirty -five years' follow-up
of the superior child. Palo Alto, CA: Stanford
University Press.
Thorndike, R. L., & Hagen, E. (1959) Ten thou¬
sand careers. New York: Wiley.
Thurstone, L. L. (1938) Primary mental abilities.
Chicago: University of Chicago Press.
Thurstone, L. L., & Thurstone, T. G. (1941) Facto¬
rial studies of intelligence. Psychometric Mono¬
graphs, 94 No. 2.
Toomey, R., Lyons, M. J., Eisen, S. A., Xian, H.,
Chantarujikapong, S., Seidman, L. J., Faraone,
S. V., & Tsuang, M. T. (2003) A twin study of
the neuropsychological consequences of stim¬
ulant abuse. Archives of General Psychiatry, 60
(3). 3°3“310-
Tucker-Drob, E., & Salthouse, T. A. (2008) Adult
age trends in the relations among cognitive
traits. Psychology and Aging, 23 (2), 453-480.
Tuddenham, R. D. (1948) Soldier intelligence in
World Wars I and II. American Psychologist, 3
W, 54-56-
Tulving, E., Habib, R., Nyberg, L., Lepage,
M., & McIntosh, A. R. (1999) Positron emis¬
sion tomography correlations in and beyond
medial temporal lobes. Hippocampus, 9 (1), 71-
82.
Turkheimer, E. (1991) Individual and group dif¬
ferences in adoption studies of IQ. Psychologi¬
cal Bulletin, 110 (3), 392-405.
482
REFERENCES
Turkheimer, E., Haley, A., Waldron, M.,
D’Onofrio, B., & Gottesman, I. (2003) Socio¬
economic status modifies heritability of IQ in
young children. Psychological Science, 14 (6),
623-628.
Van Der Maas, H. L. J., Dolan, C. V., Gras-
man, R. P. P. P, Wicherts, J. M., Huizenga, H.
M., & Raij makers, M. E. J. (2006) A dynami¬
cal model of general intelligence: The positive
manifold of intelligence by mutualism. Psy¬
chological Review, 113 (4), 842-861.
Van der Sluis, S., Derom, C., Thiery, E., Bar¬
tels, M., Polderman, T. J. C., Verhulst, F. C.,
Jacobs, N., van Gestel, S., deGeus, J. C.,
Dolan, C., Boomsma, D. I., & Posthuma, D.
(2008) Sex differences on the WISC-R in Bel¬
gium and the Netherlands. Intelligence, 36 (1),
48-67.
Van der Sluis, S., Willemsen, G., De Geus,
E. J. C., Boomsma, D. I., & Posthuma, D.
(2008) Gene-environment interaction in adult
IQ scores: Measures of past and present
environment. Behavior Genetics, 38 [4), 348-
360.
Van Ijzendoorn, M. H., Juffer, F., & Klein Poel-
huis, C. W. (2005) Adoption and cognitive
development: A meta-analytic comparison of
adopted and nonadopted children’s IQ and
school performance. Psychological Bulletin, 131
(2), 301-316.
van Leeuwen, M., van den Berg, S. M., &
Boomsma, D. I. (2008) A twin-family study
of general IQ. Learning and Individual Differ¬
ences, 18 (0, 76-88.
van Rijn, S., Aleman, A., Swaab, H., Vink, M.,
Sommer, I., & Kahn, R. S. (2008] Effects of
an extra X chromosome on language lateral¬
ization: An fMRI study with Klinefelter men
(47, XXY). Schizophrenia Research, 101 (1-3];
17— 25 ■
Vasquez, M. J. T., & Jones, J. M. (2006) Increasing
the number of psychologists of color: Public
policy issues for affirmative diversity. Ameri¬
can Psychologist, 61, 132-142.
Venter, J. C., et al. (over 200 coauthors] (2001]
The sequence of the human genome. Science,
291, 1304-1351.
Vernon, P. A., Wickett, J. C., Bazana, P. G., &
Stelmack, R. M. (2000] The neuropsychology
and neurophysiology of human intelligence.
In R. J. Sternberg (Ed.), Handbook of intelli¬
gence. New York: Cambridge University Press,
pp. 245-265.
Vernon, P. E. (1964) The structure of human abili¬
ties. London: Methuen.
Vernon, P. E. (1965) Ability factors and environ¬
mental influences. American Psychologist, 20,
723-733-
Vernon, P. E. (1979) Intelligence: Heredity and
environment. San Francisco: Freeman.
Visser, B. A., Ashton, M. C., & Vernon, P. A.
(2006) Beyond g: Putting multiple intelli¬
gences theory to the test. Intelligence, 34 (5),
487-502.
Wachs, T. D., McCabe, G. (2001) Relation of
maternal intelligence and schooling to off¬
spring nutritional intake. International Journal
of Behavioral Development, 25 (5), 444-449.
Wai, J., Lubinski, D., & Benbow, C. (2005)
Creativity and occupational accomplishments
among intellectually precocious youths: An
age 13 to 33 longitudinal study. Journal of Edu¬
cational Psychology, (97) 3, 484-492.
.Wainer, H., & Steinberg, L. S. (1992) Sex differ¬
ences in performance on the mathematics sec¬
tion of the Scholastic Aptitude Test: A bidi¬
rectional validity study. Harvard Educational
Remew, 62 (3), 323—336.
Waller, D., Beall, A. C., & Loomis, J. M. (2004)
Using virtual environments to assess direc¬
tional knowledge. Journal of Environmental
Psychology, 24 (1), 105-116.
Waller, D., Knapp, D., & Hunt, E. (2001) Spa¬
tial representations of virtual mazes: The role
of visual fidelity and individual differences.
Human Factors, 43 (i), 147-158.
Wasserman, G. A., Liu, X., Popovac, D., Factor-
Litvak, P., Line, J., Waternaux, C., Lolacono,
N., & Graziano, J. H. (2000) The Yugoslavia
prospective lead study: Contributions of pre¬
natal and postnatal lead exposure to early
intelligence. Neurotoxicology and Teratology,
22, 811-828.
Watson, J. D. (2003) DNA: The secret of life. New
York: Knopf.
Watson, J. D., & Crick, F. H. C. (1953) Molec¬
ular structure of nucleic acids: A structure
for deoxyribose nucleic acid. Nature, 4356 (25
April], 737.
Webb, R. M., Lubinsky, D., & Benbow, C. P.
(2002) Mathematically facile adolescents with
math/science aspirations. Journal of Educa¬
tional Psychology, 97, 484-492.
Wechsler, D. (1975) Intelligence defined and
undefined: A relativistic appraisal. American
Psychologist, 30 (2), 135-139.
Weinberg, R. A., Scarr, S., & Waldman, I. D.
(1992) The Minnesota Transracial Adoption
Study: A follow-up of IQ test performance
at adolescence. Intelligence, 16 (1), 117-135.
REFERENCES
483
Wetherell, J. L., Reynolds, C. A., Gatz, M.; &
Pedersen, N. L. (2002) Aging, cognitive perfor¬
mance and cognitive decline in normal aging.
Journal of Gerontology: Psychological Sciences,
57B, 246-255.
Whalley, L. J., & Deary, I. J. (2001) Longitudinal
cohort study of childhood IQ and survival up
to age 76. British Medical Journal, 322 (7 April),
i“5-
Whetzel, D. L., & McDaniel, M. A. (2006) Pre¬
diction of national wealth. Intelligence, 34 (5),
449-458.
Wicherts, J. M., Dolan, C. V., & Hessen, D. J.
(2005) Stereotype threat and group differences
in test performance: A question of measure¬
ment invariance. Journal of Personality and
Social Psychology, 89 (5), 696-716.
Wicherts, J. M., Dolan, C. V., & van der Maas,
H. L. J. (2010a) A systematic literature review
of the average IQ of Africans. Intelligence, 39
W; 1“2°-
Wicherts, J. M., Dolan, C. V., & van der Maas,
H. L. J. (2010b) The dangers of unsystematic
selection methods and the representativeness
of 46 samples of African test-takers. Intelli¬
gence, 38 (1), 30-37.
Wigdor, A. K., & Green, B. F., Jr. (1991) Perfor¬
mance assessment in the workplace. Washing¬
ton, DC: National Academy Press.
Willcut, E. G., DeFries, J. C., Pennington, B. F.,
Smith, S. D., Caron, L. R., & Olson, R. K.
(2004) Genetic eitiology of comorbid read¬
ing difficulties and ADHD. In R. Plomin,
J. C. DeFries, I. W. Craig, & P. McGuffin
(Eds.), Behavioral genetics in the postgenomic
era. Washington, DC: American Psychologi¬
cal Association, pp. 227-246.
Willerman, L., Schultz, R., Rutledge, J. N.f &
Bigler, E. D. (1991) In vivo brain size and intel¬
ligence. Intelligence, 15 (2), 223-228.
Wilson, D. S., & Wilson, E. O. (2007) Rethink¬
ing the theoretical foundations of sociobiol¬
ogy. Quarterly Review of Biology, 82, 327-348.
Wilson, D. S., & Wilson, E. O. (2008) Evolution
“for the good of the group.” American Scientist,
96, 378-389.
Wilson, W. J. (1997) When work disappears: The
world of the new urban poor. New York: Knopf.
Winship, C., & Korenman, S. (1997) Does stay¬
ing in school make you smarter? The effect
of education on IQ in The Bell Curve. In B.
Devlin, S. E. Fienberg, D. P. Resnick, & K.
Roeder (Eds.), Intelligence, genes, and success:
Scientists respond to The Bell Curve. New York:
Springer-Verlag, pp. 215-234.
Witelson, S. F., Beresh, H., & Kigar, D. L. (2006)
Intelligence and brain size in 100 postmortem
brains: Sex, lateralization and age factors.
Brain: A Journal of Neurology, 129 (2), 386-398.
Witelson, S. F., Kigar, D. L., & Harvey, T. (1999)
The exceptional brain of Albert Einstein. The
Lancet, 353 (19 June), 2150-2153.
Witkin, H. A., Mednick, S. A., Schulsinger, F.,
Bakkestrom, E., Christiansen, K. O., Good-
enough, D. R., Hirschorn, K., Lundstren, C.,
Owen, D. R., Philip, J., Rubin, D. B., & Stock¬
ing, M. (1976) Criminality in XYY and XXY
men. Science, 193 (4253) (13 August), 547-555.
Wittmann, W. W. (2005) Group differences
in intelligence and related measures. In O.
Wilhelm & R. Engle (Eds.), Handbook of
understanding and measuring intelligence. Lon¬
don: Sage, pp. 223-240.
Wolf, M. (2007) Proust and the squid: The story
and science of the reading brain. New York:
Harper.
Wonderlic Corporation (2007) Wonderlic Per¬
sonnel Test normative report. Libertyville, IL:
Wonderlic, Inc.
Wonderlic, E. F., & Wonderlic, C. F. (1992)
Wonderlic Personnel Test user's manual. Lib¬
ertyville, IL: Wonderlic Personnel Test, Inc.
Woods, R. P., Freimer, N. B., de Young, J. A.,
Fears, S. C., Sicotte, N. L., Service, S. K.,
Valentino, D. J., Toga, A. W., & Mazziotta, J.
C. (2006) Normal variants of microcephalin and
ASPM do not account for brain size variability.
Human Molecular Genetics, 15 (12), 2025-2029.
Yang, C.-F.J., Hooven, C. K., Boynes, M., Gray,
P. B., & Pope, H. G., Jr. (2007) Testosterone
levels and mental rotation in Chinese men.
Hormones and Behavior, 51 (3), 373-378.
Yeung, W. J., & Conley, D. (2008) Black- White
achievement gap and family wealth. Child
Development, 79 (2), 303-324.
Zadeh, L. A. (1965) Fuzzy sets. Information and
Control, 8, 338-353.
Zajonc, R. B. (1983) Validating the confluence
model. Psychological Bulletin, 93 (3), 457-480.
Zajonc, R. B., Marcus, H. L., & Marcus, G. B.
(1979) The birth order puzzle. Journal of Per¬
sonality and Social Psychology , 37 (8), 1325-1341.
Zerjal, T., Tyler-Smith, C., et al. (21 additional
authors) (2003) The genetic legacy of the Mon¬
gols. American Journal of Human Genetics, 72
(3)» 719_722-
Zigler, E., & Styfco, S. J. (1997) A “Head Start”
in what pursuit? IQ versus social competence
as the objective of early intervention. In B.
Devlin, S. E. Fienberg, D. P. Resnick, & K.
484
REFERENCES
Roeder (Eds.], Intelligence , genes, and success:
Scientists respond to The Bell Curve. New York:
Springer-Verlag, pp. 281-314.
Zimmerman, F. J., Chirstakis, D. A., & Melt-
zoff; A. N. (2007] Associations between media
viewing and language development in children
under age 2 years. The Journal of Pediatrics, 151
(4), 364-368.
Zuboff, S. (1988] In the age of the smart machine:
The future of work and power. New York: Basic
Books.
Zwick, R., 8c Green, J. G. (2007] New per¬
spectives on the correlation of SAT scores,
high school grades, and socioeconomic fac¬
tors. Journal of Educational Measurement, 44
0), 23-45.
Author Index
Abad, Francisco 93, 98
Ackerman, Philip 27, 130-34, 135, 136, 138, 159,
170, 171, 190, 258, 375, 433, 451
Agnoli, Franca 169
Alarcon, Marciela 239
Alfonso, Vincent 100
Alkire, Michael 190, 191
Allahyar, Maryam 60
Alpert, Nathaniel 190
Aluja-Fabregat, A. 243
Ambady, Nalini 402
Amsel, Eric 112, 334
Anastasi, Anne 292
Anderson, Daniel 283
Anderson, John R. 143
Ando, Juko 239
Ankney, C. Davison 187, 404
Arikawa, Hiroko 444
Aronson, Joshua 357, 426
Ashton, Michael 118, 364, 365
Atkinson, Richard 62, 218
Baare, Wim 433
Backhoff, Eduardo 412, 413, 414
Baddeley, Alan 44, 144, 146, 156, 170, 192, 271
Baghurst, Peter 281
Baker, Laura 243
Baltes, Paul 131
Bamshad, Michael 408, 409, 435
Barnett, W. Steven 289, 290, 296
Bar-On, Reuven 137
Batty, G. David 22, 278, 436
Bayless, Sarah 272
Bayley, Nancy 271, 272
Bazana, P. Gordon 75
Beall, Andrew 168
Beaujean, A. Alexander 239
Beir, Margeret 27, 132, 133, 159, 170, 171,
375
Benbow, Camilla 61, 346, 347, 348, 400
Berenbaum, Sheri 406
Beresh, H. 187
Berg, Cynthia 305, 390
Bernardoni, Louis 40
Bemreuter, R. G. 421
Berra, Laurence [Yogi) 448
Binet, Alfred 3, 4, 5, 7, 9, 12, 15, 32, 128, 319,
364, 423
Birnbaum, Allan 55
Bittner, Alvah 154
Bjerkedal, Tor 293, 295
Black, Paul 61
Blackburn, Carol 346, 348
Blackwell, Debra 260
Blair, Clancy 192, 300
Blinkhorn, Steven 389
Boardman, David 390
Bollen, Kenneth 222
485
486
AUTHOR INDEX
Bons, Trudy Ann 252, 412
Boomsma, Dorrit 232, 233, 240, 243, 252
Boring, Edwin G. 10, 11, 16, 17, 19, 23, 44
Born, Marise 378
Borneman, Matthew 424
Bornkessel, Ina 190
Bouchard, Tom, Jr. 68, 72, 106-109, x66, 199,
2V» 236, 237; 244> 263> 3°5> 38o> 387"88>
389, 390, 391, 403
Bowley, Graham 400
Boyle, Mary 159, 171
Bradley, Robert 287, 428
Bransford, John 298
Braver, Todd 190
Bray, Douglas 333
Broadhurst, C. Leigh 257
Brody, Nathaniel 118, 124, 137, 138, 272, 319,
346, 348
Brokaw, Tom 267
Bronfrenbrenner, Urie 219, 243, 298
Brothers, Joyce 294
Brown, Ann 298
Brown, Polly 166, 185
Brugman, Gerard 374
Brunner, Martin 391-92
Buchanan, Leo 281
Bunce, Debra 287, 288
Burgess, Neill 405
Burgess, Paul 22, 405
Burton, Lorell 166, 167, 168
Busjahn, Andreas 232
Butter worth, Brian 305
Cabeza, Roberto 190, 191, 192, 194
Campbell, Donald 403
Campbell, Prances 290, 291
Campbell, John P. 329
Caplan, David 190, 193
Caporael, Linda 275
Capron, Christiane 286
Carlson, Jerry 18, 96, 235, 358, 407
Carothers, Andrew 385
Carpenter, Patricia 74, 75, 157, 163, 195,
405
Carroll, John 79, 84-90, 91, 93, 94, 95, 99, 100,
110, 118, 163, 166, 186, 193, 200, 367
Caruso, David 136
Casey, M. Beth 397
Cattell, Raymond 34, 87, 99, 103, 104-105
Cavallo, Alexander 421
Cawley, John 421
Ceci, Stephen 58, 219, 243, 284, 298, 312, 317,
346, 360, 398, 443
Cerella, John 371, 372
Chabris, Christopher 190, 274
Chan, David 424
Chen, Jie-Qi 62, 116, 117
Chiodo, Lisa 281
Choi, Jean 390
Christakis, Dimitri 283
Christal, Raymond 157
Clark, Donald 414
Clark, Mary Jo 396
Clarke, Jane 278
Cocking, Rodney 298
Colcombe, Stanley 370, 372, 373
Cole, Jonathon 20, 50, 297, 300
Collins, W. Andrew 428
Colom, Roberto 93, 98, 153, 158, 187, 191, 193,
l94> 243; 38°; 387j 4°4
Conley, Dalton 428, 429, 432
Connelly, Brian 424
Connor, Paul 273
Contreras, Luis 412, 413, 414
Cook, Phillip 429
Cooke, Ayanna 193
Corkin, Suzanne 182
Corwyn, Robert 428
Counter, S. Allen 281
Court, John 381
Couthino, Martha 384
Coyle, Thomas 322
Craig, Ian 251, 254
Crawford, Michael 257
Crede, Marcus 136, 327
Cree, Edward 327
Crick, Francis 57, 245
Cronbach, Lee 70, 143, 349
Crouse, James 418, 419
Cullen, Michael 426
Cunnane, Stephen 257
Cvorovic, Jelena 412
Dai, Xiaoyang 378
Damasio, Antonio 137, 181
Daneman, Meredith 74, 156, 157
Daniel, Mark 98, 153, 171
Das, J. P. 36, 75, 449
De Bakey, Michael 131
de Frias, Cindy 370
De Geus, Eco 232, 233, 240, 252
Deary, Ian 12, 14, 15, 16, 76, 77, 92, 98, 142,
153-58, 184-85, 272, 278, 320, 371, 378, 379,
383, 385, 387, 436
Dehaene, Stanislas 185
Demetriou, Andreas 159
Der, Geoff 153, 272, 371
Derom, Catherine 380
Detterman, Douglas 43, 51, 56, 98, 153, 171,
291, 317, 322
AUTHOR INDEX
487
Deutsch, Gayle 193
Devlin, Bernard 350
Dewey, D. M. 421
Diamond, Jared 309, 444-45
Dick, Danielle 251
Dickens, William 244
DiLalla, Lisabeth 287, 288
Dobzhansky, Theodosius 204, 206-207
Evolution and the Origin of the Species
207
Dodrill, Carl 338
Dolan, Conor 364, 365, 380, 402, 415, 426, 439
Dombrowski, Stefan 272
Donders, Jacobus 386
Dresser, Richard 172
Below the Belt 172
Duckworth, Angela 135, 319
Duncan, John 159, 191
Duyme, Michel 286
Dweck, Carol 136, 357
Earles, James 104
Eddy, William 75
Edwards, A. W. F. 435
Edwards, Allen 137
Eichenbaum, Howard 198
El-Abbadi, Hazem 421
Elliot, Joseph 60
Elliott, Colin 92
Embretson, Susan 55
Emslie, Hazel 191
Engle, Randall 157, 158
Ericsson, Anders 98, 131, 304, 330
Ernst, Jody 231
Esbjorn, Barbara 272
Evans, Patrick 252
Eysenck, Hans 89, 235
Fabiani, Monica 370, 372, 373
Fagan, Joseph 198, 272, 431
Feldman, Ruth 344
Fernandes, Cres 92
Fernandez-Ballesteros, Roccio 439
Ferriman, Kimberly 400
Fiebach, Christian 190
Fifer, Gordon 414
Fink, Andreas 189
Finkel, Deborah 233, 240, 372, 373
Fish, Jefferson 408
Fisk, Arthur 304
Flanagan, Dawn 100
Flynn, Jim 58, 244, 265-67, 269, 348, 357,
367
Fogarty, Gerald 166, 167, 168
Forehand, Rex 287
Freebody, Peter 70
Frey, Meredith 43, 51, 56, 322
Fridjhon, Peter 412
Friederici, Angela 190, 200
Frost, Nancy 143, 149, 152
Fung, Tak 378
Furnham, Adrian 117
Gajdusek, Daniel 275
Gale, Catherine 436
Galton, Francis 2, 15, 66, 89, 142, 145, 186, 189,
217, 219, 255, 292, 309
Hereditary Genius 217
Ganis, Georgio 185, 194
Garcia, Luis 93
Garcia-Lopez, Oscar 243
Gardner, Howard 4, 61, 113-19, 120, 131, 138,
342, 343, 449
Frames of Mind 114
Garlick, Dennis 198
Geary, David 47, 48, 70, 186, 369, 403, 444
Geiser, Saul 62, 323, 327
Gelade, Garry 384, 394
Gentner, Dedre 70
Gentner, Donald 70
Georgiou, George 278
Gigerenzer, Gerd 112, 374
Gignac, Giles 91, 95, 155, 163
Gilbert, William 203
Princess Ida 203
Goff, Maynard 134, 135
Goldberg, Robert 163
Goldman, Samuel 271
Goleman, Daniel 137
Goodwin, Doris 61
Gopnik, Alison 271
Gorey, Kevin 289
Gosso, M. F. 251
Gottfredson, Linda 1, 16, 22, 95, 124, 126, 130,
132, 270, 329, 332, 337, 338, 340-42, 358
Gotz, Michael 251
Gould, Stephen 185, 186
Grafton, Frances 98
Grandy, Jerilee 396, 398
Gray, Jeremy 189
Green, Bert 330
Green, Jennifer 273-84, 320
Greenfield, Patricia 308
Grewal, Daisy 136
Grigorenko, Elena 120-36, 307, 332, 358
Grisssmer, David 417
Gronbaek, Moren 278
Guiso, Luigi 401
Gunnarsdottir, Eyrin 321
Gur, Raquel 405
488
AUTHOR INDEX
Gur, Ruben 405
Gustafsson; Jan-Erik 95, 102, 103, 104-105
Habib; Reza 198
Hagen, Elizabeth 41, 334—36^ 338^ 341
Haier, Richard 37, 137^ 176-92, 198,
4° 5
Hakstian, A. Ralph 103, 419
Halpern, Diane 109, 377, 398, 400, 406
Hamaker, Ellen 364, 365, 381-400, 415
Hamilton, Alice 280
Harden, K. Paige 217, 232, 243
Hardison, Chaitra 426
Harford, Thomas 278
Harris, John 271
Hartigan, John 424
Hartley, John 372
Hartmann, Peter 414
Harvey, John 292
Harvey, Thomas 185
Hattori, Kanetoshi 226, 378
Haughton, Helen 278
Hausmann, Markus 406
Hayes, Brian 104
Hazlett, Heather 251
Hearnshaw, Leslie 234, 235
Hedges, Larry 387, 392, 418
Hedlund, Jennifer 126
Heeb, Randal 421
Hegarty, Mary 48, 60, 70, 169, 170
Heggestad, Eric 131, 132
Heitz, Richard 158
Helms-Lorenz, Michelle 93, 414, 415, 425
Herrnstein, Richard 8, 9, 16, 22, 104, 113, 265,
269, 284, 289, 291, 292, 298, 299, 319, 320,
323, 350, 384, 412, 420
The Bell Curve 311
Hertzog, Christopher 305, 390
Hessen, David 402, 426
Hezlett, Sarah 328, 333
Hill, Linzey 252
Hill, Nicole 304
Hitch, Graham 156
Hoeft, Fumiko 193
Holland, Cynthia 198, 272, 431
Holland, James 131
Holohan, Carole 344
Holt, Matthew 37
Horn, John 2, 34, 87, 99, 103, 105, 109, 110,
370
Howard, Ann 333
Hoyer, William 372
Huarte de San Juan, Juan 2, 15, 34
Hubbs-Tait, Laura 279, 281
Humphreys, Lloyd 32, 48, 380
Hunt, Earl 18, 55, 60, 61, 77, 98, 108, 112, 116,
124, 128, 135, 143-69, 195, 235, 300, 302, 305,
332/ 336. 34b 358. 359. 362. 368. 389. 39°.
395. 4°5. 4°7. 439. 44°
Hunt, Steven 119, 122, 330
Hunter, I. M. L 305
Hunter, John 128, 135, 317, 318, 330, 331, 332,
333
Hutchins, Edward 124, 297
Hyde, Janet 401
Inlow, Jennifer 246
Irwing, Paul 96, 302, 303, 378, 381, 382, 396,
397. 4°2
Jackson, Douglas 32, 33, 359
Jacoby, Larry 105, 196
Jaeggi, Susanne 306
Jarvik, Lissy 250
Jensen, Arthur 46, 89, 90, 91, 95, 100, 103,
104-105, 132, 142, 145, 150, 153, 200, 234-35,
239, 289, 291, 304, 319, 338, 339, 355, 363,
364> 365> 379/ 38o> 412/ 414"15- 425> 433~34
Johannesson, Hinrik 321
Johnson, Ray 40
Johnson, Wendy 72, 94, 95, 106-109, *66, 199,
219, 226, 233, 235, 238-40, 263, 305, 380,
385, 387-88, 389, 390, 391
Johnstone, E. C. 251
Jones, Ben 278
Jones, James 349
Jones, Stephanie 274
Jonides, John 190, 192
Joreskog, Karl 84, 85
Juan-Espinosa, Manuel 21, 93, 370
Juffer, Femmie 285
Jung, Rex 37, 137, 188, 191-92, 405
Jungeblut, Ann 301
Just, Marcel 48, 70, 74, 75, 163-64, 170, 189,
!93, 195, 405
Kahn, Ashraf 278
Kamin, Leon 234
Kanaya, Tomoe 57, 58
Kanazawa, Satsoi
Kane, Michael 157, 158
Kaplan, Charles 320
Karama, S. 187
Katigbak, Marcia 288
Keller, Timothy 74, 75, 193
Kennedy, Joanna 251, 254
Kieras, David 143, 145
Kigar, D.L. 185, 187
AUTHOR INDEX
489
Kimura, Doreen 389, 405, 406
Kintsch, Walter 74, 161, 164
Kirby, John 36
Kirkham, D. I. 270
Klein, Gary 129, 374
Klodin, Victor 250
Knapp, David 169, 390
Knapp, Deirdre 329
Knight, Robert 192
Kobrin, Jennifer 323
Koenig, Katherine 51, 56, 322
Korenman, Sandor 299
Kornhaber, Mindy 4
Kornhauser, Arthur 284
Kosslyn, Steven 166, 167, 194
Kotchick, Beth 287
Kovarik, William 281
Kovas, Marta 240, 242
Kramer, Arthur 370, 372, 373
Krampe, Ralf 304
Krauss, Stefan 391
Kremen, William 239
Kristensen, Petter 293, 295
Kroc, Rick 323
Kronman, Anthony 164-65
Kruuse, Nanna 414
Kuhl, Patricia 271
Kuncel, Nathan 136, 327, 328, 333,
342
Kunter, Marelke 391
Ky, K. N. 274
Kyllonen, Patrick 157
Laland, Kevin 185
Lange, Rael 383
Langer, Rosanna 112, 171
Lansman, Marcy 143
Larson, Gerald 190
Laterbauer, Paul 177
Lave, Jean 300
Law, David 389
Leckliter, Ingrid 278
Lee, H. 193
Lee, Kibeom 364, 365
Legree, Peter 98
Lemann, Nicholas 31, 42, 113, 217,
322
Lesser, Gerald 414
Levidow, Bjorn 303
Lewis, Charles 396
Lewis, Joe 149
Lewontin, Richard 287, 408
Lichter, Daniel 260
Lidz, Carol 60
Liker, Jeffrey 48, 312, 317
Lindquist, Everett 5
Lindzey, Gardner 410, 411
Lippmann, Walter 10, 16, 17, 19, 23
Loehlin, John 217, 223, 231, 243, 269, 410,
411
Logie, Robert 170
Lohman, David 41
Longman, R. Stewart 378
Loomis, Jack 168
Loutzenhiser, Lynn 112, 334
Lubinski, David 32, 48, 61, 346, 347, 348, 380,
400
Luciano, Michelle 238, 239, 254
Luders, Eileen 187
Ludwig, Jens 429
Lunneborg, Clifford 143-44; 149; 152
Luria, A.R. 37, 75
Lynn, Richard 226, 269, 292, 355, 356, 378-79,
38^82, 387, 396, 397, 402, 404, 412, 413,
4H> 4*7; 433-34; 436; 437-41; 444 443"45;
446
MacDonald, Maryellen 74
Mace, Ruth 294
Machiavelli, Niccolo 22
The Prince 22
Machin, Stephen 393, 401
Macintosh, Nick 235, 320, 404
Macintyre, Sally 278
MacLeod, Colin 145, 195
Maddox, B. 51
Madhyastha, Tara 341, 359, 366, 395
Maguire, Eleanor 195, 405
Malinowski, J. C. 390
Mansfield, Peter 177
Margret, M. 373
Markus, Gregory 293
Markus, Hazel 293
Marsh, Laura 196, 198
Marsiske, Michael 373
Martin, Roy 272
Matarazzo, Joseph 12, 32, 278, 319,
378
Mathews, Nancy 145, 195
Matsuyama, Steven 250
Matte, Thomas 272
Mattern, Krista 420-24
Matthews, Gerald 138
Mau, Wei-Cheng 396, 397
Max, Jeffery 192
Mayer, John 120-36
McArdle, John 368, 369
McAuley, Elizabeth 251
49°
AUTHOR INDEX
McCabe, George 274
McCamley-Jenkins, Laura 396
McDaniel, Michael 187, 333, 420, 440
McGrew, Kevin 100, 101, 102, 103, 368
McGue, Matt 20, 227, 232, 237, 277
McHenry, Jeffrey 330
McIntosh, Anthony 198
McKay, Patrick 420
Mekel-Bobrov, Nitzan 252
Meltzoff, Andrew 271, 283
Merikle, Philip 74, 156
Mervis, Carolyn 161
Messick, Samuel 301
Meyer, David 143, 145
Milner, Brenda 182
Minstrell, Jim 61, 300, 302, 304
Mischel, Walter 135
Moeser, Shannon 168
Moffat, Scott 195
Molenaar, Dylan 380
Molfese, Victoria 287, 288
Morrison, Ian 48, 70, 170, 288
Mortensen, Laust 278
Moselhy, Hamdy
Muller, Carol 349, 399, 400
Murray, Charles 8, 9, 16, 22, 104-105, 113, 265,
269, 283, 284, 289, 291, 292, 298, 299, 309,
319, 320, 323, 341, 342, 350, 355, 375, 384,
397, 412, 417, 420
Myerson, Joel 372
Naglieri, Jack 36, 37
Nathawat, S. S. 226
Needleman, Herbert 279
Neisser, Ulrich 16, 265
Nettlebeck, Ted 152
Neubauer, Aljoscha 75, 189, 190
Newcombe, Norah 272, 390
Newell, Allen 143
Nguyen, Hannah-Hanh 426
Nisbett, Richard 16, 50, 217, 220, 231, 242, 258,
260, 289, 294, 433, 434
Noll, Cynthia 2, 34, 99, 367
Noonan, Kelly 272
Nowell, Amy 387, 392, 418
Nyberg, Lars 190, 192
Nyborg, Helmut 191, 194, 338, 339, 364, 380,
414
O' Grady, Kevin 155
O’Keefe, John 405
O’Toole, Brian 270
Oberauer, Klaus 158
Oden, Melita 344, 345
Ogbu, John 429, 431, 432
Ohnishi, Takashi 405
Okazaki, Sumie 357, 431
Ones, Deniz 328, 333
Ono, Yutaka 239
Opitz, Bertram 200
Ortega, Fernando 281
Oswald, Donald 384
Otis, Arthur 40
Owen, David 266, 284, 288
Pal, Shish 226
Palmer, John 48, 151, 160, 163, 380
Papierno, Paul 346
Park, Gregory 348, 400
Parker, Douglas 278
Parker, Elizabeth 278
Payne-Gaposchkin, Cecilia 397
Peake, Philip 135
Pearson, Karl 186
Pedersen, Nancy 233, 240, 243, 372, 373
Pekkarinen, Tuomas 393, 401
Pelligrino, James 389
Peltonen, Leena 232
Peters, Laurence 298
Petrill, Stephen 230, 231, 232
Phillips, Meredith 418, 419, 430
Piazza, Manuela 185
Pifer, Mark 98
Pillow, David 322
Pind, Jorgen 321, 322-23
Pittinsky, Todd 402
Plomin, Robert 219, 220, 223, 227, 231, 237, 251,
252, 254
Plucker, Jonathon 122
Poelhuis, Caroline 285
Polderman, Tinea 238, 239
Pollitt, Ernesto 277
Poltrock, Steven 155, 166, 185
Poortinga, Ypes 93, 414, 415, 425
Porteus, Stanley 421
Posner, Michael 148, 149, 183, 192, 306,
3°7
Posthuma, Danielle 232, 233, 240, 252, 254
Prat, Chantel 189, 193
Preston, Samuel 269
Pulakos, Elaine 424
Puri, Prena 226
Puts, David 406
Rabi, Isidor 177
Radwan, Susan 100, 359
Raftery, Adrian 108
Ralph, John 418
AUTHOR INDEX
491
Ramist, Leonard 396
Ratcliffe, S. G. 231
Rauscher, Frances 274
Raven, Jeanne 46, 382
Raven, John 46, 56, 269, 284, 329, 333, 381-400,
411, 413, 414, 415
Raven, John C. 46, 95, 266, 413
Raz, Naftali 370, 373
Reader, Simon 185
Ree, Malcom 104
Reich, Robert 311, 336
Reichle, Erik 195, 197, 405
Reise, Steven 55
Resnick, Susan 406
Restifo, Linda 248
Reynolds, Chandra 236, 240, 243
Rindermann, Heiner 437, 441-43, 445
Robb, George 40
Roberts, Richard 104, 138, 300, 322
Robinson, Halbert 349
Robinson, Nancy 299, 349
Rodgers, Joseph 293, 294
Rohde, Treena Eileen 33
Rolfhus, Eric 133
Roorda, Willemijn 364, 365
Rosen, Allyson 198
Ross, Judith 249
Roth, Philip 420
Rothbart, Mary 306, 307
Rowe, Richard 243
Rushton, J. Phillipe 32, 33, 186, 187, 252, 355,
359, 364, 412, 415, 433-34, 446
Ryan, Ann Marie 426
Sackett, Paul 284, 323, 325, 327, 360, 424,
426
Saklofske, Donald 98, 378
Salovey, Peter 136
Salthouse, Timothy 153, 171, 370, 372,
373
Santy, Patricia 354
Sarason, Irwin 113
Scarr, Sandra 227, 231, 285, 296
Schaie, K. Warner 261-63, 2^4, 265, 318, 369,
37 0/ 37b 374» 375
Schaler, Jeffery 117
Schiff, Michel 287
Schmidt, Frank 128, 135, 317, 330, 331, 332,
333
Schmitt, Neal 424
Schnakenberg, Robert 163
Schneider, Walter 304
Schooler, Carmi 308
Schottenbauer, Michelle 278
Schrauseer, Dietmar 189
Schwartz, Steven 163
Scullin, Matthew 58, 336, 337
Sears, Robert 344
Segal, Nancy 227
Seligman, Martin 135, 320
Shakespeare, William 1, 47, 137, 446
Comedy of Errors 1
C oriolanus 257
King Henry V 137
Shallice, Tim 191
Shapiro, Fred 64, 95, 376
Shaw, G. 274
Shaw, P. 199
Shenkin, Susan 272
Sheppard, Leah 433
Shiffrin, Richard 304
Shih, Margaret 402
Shimamura, Arthur 192
Shoda, Yuichi 135
Shute, Valerie 70
Shyam, Radhey 226
Sigman, Marian 276
Silverman, Irwin 390
Simon, Theophile 5, 32
Simonton, Dean 61, 121-22, 185, 319, 343
Singh, Rajbir 226
Skarbrevik, Karl 384
Skeels, Harold 284
Skodak, Marie 284
Skuy, Mervyn 412
Smedley, Audrey 408, 409
Smedley, Brian 37, 408, 409
Smith, Edward 190, 192
Smith, Glen A. 239
Smith, Jacqui 131
Smith, Pauline 92, 378, 383, 387
Smith, R. W. 270
Snow, Richard 51, 70, 297-98, 347
Snow, William 378
Sorboom, Dag 85
Sorenson, Thorkild
Spearman, Charles 34-36, 46, 83-89, 90, 91,
92, 95, 98, 100, 109, 300, 363, 407
Spry, Kathleen 317
Spuhler, J. N. 410, 411
Staaudinger, Ursula 131
Stankov, Lazar 99, 103, 166
Stanley, Julian 346
Steele, Claude 357, 426
Stein, Zena 276
Steinberg, Laurence 428
Steinberg, Linda 396
Stelmack, Robert 75
492
AUTHOR INDEX
Stephens, Deborah 157
Stern, W. 7
Sternberg, Robert J. 2, 27, 60, 112, 113, 120-30,
131, 138, 140, 155, 301, 306, 307, 332, 358,
373- 374- 448/ 449, 451
Practical Intelligence in Everyday Life 121,
124, 130
Sternberg, Saul 149
Stevenson, Jim 272
Stewart, Manard 163
Sticht, Thomas 48, 350, 351, 352
Stoolmiller, Mike 232
Strand, Steve 92, 320, 378, 383, 387
Streissguth, Anne 273
Strieker, Larry 402, 426
Strother, Charles 261, 262, 264
Studley, Roger 62, 323, 327
Styfco, Sally 289
Sue, Stanley 357, 431, 432
Sullivan, Arthur 203
Sullivan, Edith 196, 198
Sunderland, Alan 271
Suzuki, Lisa 426
Tan, Sherylle 406
Tang, Hua 408, 435
Tannenbaum, Abraham 1
Te Nijenhuis, Jan 107, 302, 414, 426
Teasdale, Thomas 266, 267, 284
Templer, Donald 444
Terman, Lewis 10, 12, 16, 344, 345
Tesch-Romer, Clemens 304
Teuber, Hans 182
Thompson, Leann 33, 291
Thomson, Godfrey 14
Thorndike, Robert 41, 334-36, 338,
W
Thurstone, L. L. 97-98
Thurstone, T. G. 97
Tirre, William 70
Toomey, Rosemary 278
Torrance, E. Paul 122, 125
Toth, Jeffery 105, 196
Tucker-Drob, Eliot 370
Tuddenham, Read 261, 263, 264, 268
Tulsky, David 386
Tulving, Endel 198
Turkheimer, Erik 217, 232, 243, 286
Unsworth, Nash 158
Van Court, Marian 292
Van de Vijver, Fons 93, 414, 415, 425
Van den Berg, Stephanie 243
Van der Flier, H. 302
Van der Maas, Han 97, 439
Van der Sluis, Sophie 233, 380
Van Ijzendoorn, Marinus 285
Van Leeuwen, Marieke 243
Van Rijn, Sophie 251
VanVianen, Annelies 302
V anhanen,Tutu 355, 417, 436, 437-41, 442, 443,
445
Vasquez, Melba 349
Verhaeghen, Paul 372
Vernon, Philip A. 75, 118, 252, 433
Vernon, Philip E. 89, 106
Visser, Beth 117-18, 119
Von Croman, D. Yves 200
Wachs, Theodore 274
Wai, Jonathon 400
Wainer, Howard 396
Wake, Warren 4
Waldman, Irwin 285
Waller, David 60, 168, 169, 384
Ward, William 402, 426
Warner, Molly 338
Wasserman, Gail 281, 288
Waters, Gloria 190
Watson, James 245
Wechsler, David 32, 39, 71, 89
Weidl, Kurt 96
Weinberg, Richard 231, 285
Werder, J. K. 368
Wetherell, Julie 375
Whaley, Shannon 276
Whalley, Lawrence 14, 15, 16
Wheeler, Karyn 198, 272
Whetzel, Deborah 440
White, Nathan 190, 191
Wicherts, Jelte 364, 365, 380, 402, 426,
459
Wicket, John 75
Wigdor, Alexandra 330, 424
Wiliam, Dylan 61
Willcut, Erik 242
Willemsen, Gonneke 233
Willerman, Lee 187, 231
Williams, Wendy 284, 298, 360, 398
Wilson, David 403
Wilson, Edward 403
Wilson, William Julius 430, 432
Winship, Christopher 299
Witelson, S. F. 185, 187
Witkin, Herman 251
Wittmann, Werner 160, 439, 440,
442
Wolf, Maryanne 20, 27, 428
Wonderlic, E. F. 44
Woodcock, Robert 102, 103, 368
Wright, Margaret 238, 239
Yang, C. F. 406
Yeung, W. Jean 428, 429, 432
Yonelinas, Andrew 105, 196
Zajonc, Robert 293
Zeidner, Moshe 138
AUTHOR INDEX
Zerjal, Tatiana 269
Zhang, Li-Fang 120, 124, 129, 307
Zhu, Jianjun 386
Zigas, Vincent 275
Zigler, Edward 274, 289
Zimmerman, Frederick 283
Zinn, Andrew 249
Zuboff, Soshanna 336
Zwick, Rebecca 284, 320
493
Subject Index
Abacus 96, 302, 303
ABC Deri an project 290-91, 295, 296, 310
Ability 60
auditory (Ga) 103, 108, 166
intellectual 369
learning 60
reflecting 61
Academic achievement 21, 22, 68, 90, 91, 122,
127, 135, 142, 220, 235, 242, 266, 277, 284,
289-9!, 3J9, 320, 321, 324, 325, 336, 339, 345,
346, 347, 355, 396, 401> 410, 424, 426, 428,
431, 442, 447, 450
Academics 8, 14, 17, 240
core courses in 320
practical topics 320
Accomplishments (nonacademic) 450
Asian-Americans 431
gifted 336, 346, 347, 355
Accuracy of tests 14
ACE model 220, 225-26
Adams, John (US President) 31, 218
Adams, John Quincy (US President) 218
Adenauer, Konrad (Chancellor, West
Germany) 364
Admission decisions (college or
university) 326-27
Adolescent Adoption study 231
Adoption, adoptees 69, 223, 228-31, 232,
284-87
Affirmative action (social policy) 410
AFQT score categories I-IV, defined 350
Africa 18, 255, 361, 408, 416, 446
sub-Saharan 439, 442, 443, 445
African American (Black) 53, 231, 235, 272,
285, 288, 290, 336, 339, 345, 348, 357-59,
361, 363, 364, 404, 408, 409, 410, 411-12,
4i5-i9, 420, 421, 423, 424, 425, 426, 427, 428,
429-30, 431, 432, 433, 434
community bifurcation 430
Age, age groups 131, 183
adolescent, teenage 20, 96, 132, 210, 228, 232,
284, 285, 308, 319, 336, 347, 378, 381, 382,
387, 39i, 4°i, 4°3, 4°4, 4!7, 428, 441
adulthood 49, 206, 228, 232, 237, 238, 272—79,
290, 304, 336, 366, 368, 374, 378, 381, 387,
433
childhood 132, 232, 238, 240, 248, 272, 275,
297, 321, 378, 382
chronological 5, 6
infancy 240, 248, 270, 271, 272, 275, 283
mental 5, 6
middle-aged 210, 239, 248, 305, 346, 370
middle-school 346, 378
old, elderly 20, 96, 171, 176, 206, 228, 237,
240, 243, 249, 367, 371, 373, 375, 450
pre-school (2-5) 272, 285, 306
schoolchildren 305, 308, 364, 412, 418, 441
twenties, young adult 367, 368, 371, 387
494
SUBJECT INDEX
495
Age differences in intelligence 357, 359
Age requirements for occupations 356
Aging 24; 105, 106, 131, 132, 304, 364, 445
Air Force (study of cadets) 334, 338
Aitken, Alexander 305
Alcohol, alcoholism 15, 20, 196, 211, 219, 243,
258, 273, 277-79, 307, 432, 450
Alexander the Great 297
Allele, definition of 245
Alzheimer’s disease 2, 20, 196, 209, 237,
249-50, 251, 258, 364, 375
American College Test (ACT) 50-51, 126, 322,
326-27
American Psychological Association 16
Amerindian (Native American) 50, 273, 409,
412, 413
Amnesia 270
Amphetamines 307
Amygdala 174, 182
Analytic (academic) intelligence 121, 125
Anxiety 375
Aphasia 176
Apolipoprotein E (APOE) gene 249
Aptitude 298
Aptitude x treatment interaction 70
Aptitude-achievement distinction 50-51
AR gene 213
Arithmetic 91, 153, 305
Armed Forces Qualifying Test (AFQT) 2, 8,
16, 21, 43, 56, 73, 104, 121, 128, 292, 299,
319, 322, 323, 330, 336, 337, 350, 378, 379,
420
Armed Services Vocational Aptitude Battery
(ASVAB) 23, 33, 39, 43-44, 48, 56,
103-104, 128, 157, 292, 299, 322, 329, 330,
350, 352, 378, 379, 380
Army Alpha Test 10, 13, 32, 41, 262, 263, 448
Army Beta Test 13
Army General Classification Test 262
Asian (as racial/ethnic group) 18, 348-49, 361,
408, 412, 433, 434, 443
East Asians 439, 446
Southeast Asians 439
Asian women 402
Asian-American 232, 357, 408, 410, 411, 414, 415,
421-22, 424, 425, 426, 427, 428, 429, 431, 432
ASPM gene 252
Assortative mating 243, 269
Astronauts 354
Attention 21, 77, 84, 96, 109, 110-71, 199, 240,
249/ 272/ 274/ 276> 277> 278/ 3°4/ 3°6/ 3°8/
370, 372, 374, 433, 445, 449, 450
Attention disorder 192
Attrition, from research study 367
Attrition rate, from training programs 351, 352
Autoimmune disease syndrome (AIDS) 275
Automation of cognitive actions 304, 305
Autosomes, definition of 212
Aviation officers, cadets 334, 335
Baby Einstein (learning program) 257, 283
Baldness, male pattern 213
Bartlett’s Familiar Quotations 356
Bayes’ law 108
Bayesian Information Criterion (BIC) 108
Bayley scales of development 271
Behavioral inventory 136
Bell Curve, distribution of intelligence 9
Bell Telephone System study 333
Bellingham (WA) Herald 73
Berlin Wisdom Theory 374
Binge drinking 279
Biographic data 329
Birth order 293-94, 295
Birth weight 272-73
Black (racial/ethnic group), see African
American
Blackboard model of cognition 142-43, 144,
!55/ j56
Blood oxygen level dependent (BOLD)
response 178, 179
Blood pressure, see cardiovascular disease
Body size 183-86
Bohr, Niels 448
Brain 15, 20, 21, 24, 36, 71, 74-75, 76, 77, 84,
110-202
activation of areas 189, 190, 193, 405
damage 22, 26, 75, 137, 140, 173, 181, 191, 204,
249, 260, 270-71, 366, 405
density 187, 188, 191, 193
development 405, 446
metabolism 173, 190, 449
size (cortical volume, mass, etc.) 173,
183-85, 187, 188, 201, 252, 370, 404, 433-34,
444; 446, 449
structure 174, 176, 177, 180, 187, 398, 404,
446
Brigham, Carl 41
British Ability Scale 92
Broca, Paul 176
Broca’s Area 75, 175, 193
Brunswick, Egon 160
Brunswickian symmetry 160, 193, 258
Burt, Cyril 89, 234-35, 236
Bush, George H. W. (US President) 172, 218
Bush, George W. (US President) 218, 319,
343
Cannibalism 275
Cardiovascular disease 375, 409
496
SUBJECT INDEX
Careers 397, 400
Carroll, Lewis (Charles Dodgson) 305
“The Walrus and the Carpenter” 305
Case studies 184, 185, 187
Category identification 148
Category IV (military recruits) 350-53
Category V (military recruits) 350
Catherine II (“The Great,” Empress of
Russia) 376
Cattell, James McK. 142
Causal modeling 360
Cerebellum 174
Cerebral cortex 26, 37
Cervantes, Miguel 47
Challenge hypothesis 24, 26, 307, 334, 431
Challenges, strategies for dealing with 307
Chauncey, Henry 41-42, 50
Children 9
Choice reaction time (CRT) 147, 150, 151, 153,
238> 239« 371
CHRM2 gene 251-52
Chromosomes 212, 245-46, 250
sex chromosomes 246
X chromosome 213, 246, 248, 250
Y chromosome 213, 246-47, 250
Church, Timothy 288
Cingulate gyrus, cortex 174, 182, 188, 198, 200,
306, 353, 372
anterior cingulate 175, 180, 190, 191-92, 199,
219
Civil Workplace 330
Clinton, Hillary (US Senator) 162
Clinton, William (US President) 95
Closed head injury, see concussion
Closure of forms (CS) 166
Closure task 169
Cluster analysis 386
Coaching 62
Coding (subtest of WAIS) 91
Codon, definition of 245
Cognition 22, 274
Cognitive Abilities Test (CAT) 41, 92, 320,
37 8> 383
Cognitive Assessment System (CAS) 36-38
Cognitive psychology, approach to
intelligence 141, 142
Cohort effects (Flynn effect) 105, 261, 296,
309, 366, 367, 434
design issues 268
international data on 265
Cohort-sequential study 262, 367
College (university, tertiary education) 2, 117,
130
enrollment rates 438, 440
College Board 39, 323, 326
College students (university students) 97, 133,
H8, U2> 153/ 157/ !58/ !59z Uk 189. *95/ 239,
304, 306, 359, 410
Collinearity 258, 260-61, 270, 272, 278, 279-81,
283, 319, 360
Colorado adoption study 230, 237, 239
Competence
cognitive 82-83, 255, 378
legal 13
mental 1, 9, 12, 116, 130, 255
Computer-controlled testing 39
Computerized axial tomography (CAT) 177,
179
Conant, James Bryan 41, 62
Concussion 271
Confluence model 293
Confucius 4, 356
Congenital adrenal hyperplasia (CAH) 405
Congruence (statistical condition) 365
Conscientiousness 22, 133, 135, 330
Contingent item presentation 34-38
Continuous traits 209
Corpus callosum 174
Correlated vectors 363, 365, 379, 414-15, 425,
434
Cost of testing 13, 39
Crazy Horse 3
Creativity, creative intelligence 61, 121, 122,
125, 128, 138, 374
Cree, Edward (British Naval Surgeon) 327
Cross-sectional studies 105, 133, 261, 366,
367
Cruetzfeldt- Jakob disease 275
Crystallized intelligence (Gc) 34, 99, 102-106,
108, 110, 128, 129, 130, 131, 132, 133, 134, 135,
191, 193, 196, 267, 268, 271, 297, 298, 299,
301, 322, 363, 367, 368, 369, 370, 371, 373,
374, 378, 417, 428, 446
Culture, cultural 415, 424-25, 426, 430
Curie, Marie 177
Custer, George (US General) 3
Darwin, Charles 2, 64-65, 283, 451
Decision time 150
De differentiation hypothesis 370
Delphic Oracle 65
Dementia 375
Democratization, index of 438
Department of Defense 9, 13, 43, 329, 350
Department of Labor 14, 23, 387, 392
Depression (of 1930s) 354
Deviation score, d (between groups) 361, 364,
365. 378
Dictionary of Occupational Titles (DOT) 340
Diet (dietary) 275
SUBJECT INDEX
497
Differential Aptitude Battery (DAT] 370, 387
Differential impact 52-53
Differential item functioning 362
Diffusion tensor imaging 177, 198
Digit span 155
backward 155
forward 155
Digit Symbol (subtest of WAIS) 91
Discrimination, social (prejudice) 355, 399,
409, 410, 447
Distal influences 68, 215
Distal-proximal distinction 215, 258-60
Distribution
normal or Gaussian 53, 54
of test scores 53
Divergent thinking 122
Diversity, racial/ethnic 327
DNA (deoxyribose nucleic acid) 245, 252,
253-54, 416
junk DNA 246, 247, 344
Dominant trait, definition of 205, 210
Donders, Franciscus 145-47, *48/ H9> 151, *79
Dorso-lateral prefrontal cortex (DLPFC) 175,
191
"Drop in from the Sky" testing paradigm 9,
12f 24> 47' 59' 6i; 97, 110, 139, 195, 240, 449,
45i
Du Bois, William E. B. 411
Dual tasks 157
Dynamic testing 60-61
Dysgenic effect 269, 292
Dyslexia 48
Early childhood intervention 288-91
Economic freedom, index of 442
Economic growth, index of 442
Economic success, see income
Education (schooling) 1, 2, 4, 27, 48, 68, 70,
114, 120, 231, 241, 260, 283, 284, 287,
296-300, 301, 302, 309, 337, 342, 351, 360,
386, 437, 439, 441, 443, 445, 450
male-female differences in 391
special education 241, 322
Educational achievement 16, 397, 440
Educational Testing Service (ETS) 9, 32, 42,
104, 301, 387
Efficiency of brain or nervous system 189-90,
200
Egyptian scribes 297
Egyptians 3
Einstein, Albert 2, 9, 61, 64, 65, 121, 184-85,
2°2' 343' 446' 451
Eisenhower, Dwight (US President) 2, 176
Electroencephalogram (EEG) 75, 178, 179,
180, 189
Elementary cognitive tasks (see also
information processing) 141, 142-52, 154,
433
Elementary (primary) school 29, 117, 303, 319,
4l8' 427
Eliot, T. S. 343
Elizabeth (“of Russia,” Empress of Russia) 376
Elizabeth I (Queen of England) 376
Emotional intelligence 136-38
Encephalization ratio 183, 185
Endogenous variable 28
Entrance examinations (college,
university) 324-25
Environment, environmental influences 123,
229, 230, 269, 309
family 28, 29, 295
physical 20, 24, 26, 257, 260, 270, 450
social 20, 257, 260, 283-96, 307, 308, 450
Environmental effects 242, 243, 256, 257, 258,
349
nonshared, definition of 225, 226, 231, 237,
240, 243, 291
shared, definition of 225, 226, 240
Epistasis 219
Equatorial Guinea 21
Ethnic groups, ethnicity 18, 255, 283, 351
Europe, medieval 3
European 18, 255, 408, 413, 439
Event-related potential (ERP) 178, 179,
189
Evolution 2, 22, 183, 204, 206, 207, 208, 219,
283, 416, 451
effects due to sex roles, “man the
hunter ...” 402, 432
prehistoric Homo sapiens 403
social 444
Exchange of ideas 444, 445
Executive processes, processing 144, 156, 171,
l83. 37°. 372» 374
Exogenous variable 28
Experience 330, 373, 374
Expert performance 131, 304, 317, 375
Factor analysis 221
Factor scores 379
Factors (psychometric) 72, 104
group 90
hierarchical model of 90
language 90
second-order 90
third-order 90
visual-spatial 90
Family size 292-94, 296, 310, 442
Famine 26, 28
Fang (African tribe) 21
498
SUBJECT INDEX
Faust, Drew Gilpin 397, 398
Fertility 269, 442
Fetal alcohol syndrome 20, 273
Field independence task 389
Fluid intelligence (Gf) 34, 99, 102-106, 108,
109, 110, 131, 132, 133, 158, 159, 191, 267, 270,
276 , 278, 283, 297, 298, 301, 302, 322, 363,
367, 368, 369, 370, 371, 372, 373, 374, 417,
446
improvement of 302
Ford, Gerald (US President) 1-2, 172
Ford, Henry 136
Fornix 182
FOXP2 gene, 242
Fragile X syndrome 248
Franklin, Benjamin 280
Frontal cortex, frontal lobe, forebrain 20, 37,
70, 75, 137, 174, 176, 180, 181, 182, 188, 190,
l9lt 193_94/ *95/ 196, 198, *99/ 200, 2°2, 249/
258, 270, 278, 306, 353, 370, 372, 373, 449
Full scale IQ (FSIQ) 32, 34, 188, 200
Functional magnetic resonance imaging
(fMRI) 178, 179, 189, 190, 195, 243, 251
Fuzzy concept, definition of 408
Gage, Phineas 181, 184, 191-92
Gandhi, Mahatma 61, 343
Gauss, Carl Fredrich 27
Gender gap index 401
Gene-by-environment correlation 217-18, 219,
228, 243
Gene-by-environment interactions 219, 221,
228, 229, 243, 244
General Aptitude Test Battery (GATB) 14,
340, 424
General intelligence (g) 33, 70, 72, 75, 89-99,
102, 104, 107, 108, 109, 115, 131, 135, 138, 145,
155/ U8, U9/ 16°/ l63/ i64/ U0/ i83/ i87/
188, 191, 192, 200, 231, 236, 238, 242, 263,
270, 272, 276, 278, 297, 320, 322, 328, 330,
33b 332/ 338/ 34b 363-64/ 365/ 367/ 368;
369/ 37°/ 372, 378-4U/ 4H/ 4U/ 425/ 446/
449/ 450
correlated traits hypothesis for 94, 451
sex differences on 377, 378-82, 386,
389
unitary hypothesis for 94
Generalization of results 315, 359
Genetic correlation 236
Genetic marker 253
Genetic pathologies 247-51
Genetic variation 20, 109, 199, 203—56, 416
between racial/ethnic groups 408, 45
international 439
within racial/ethnic groups 408
Genetics, genetic influences 19, 24, 25, 28, 68,
202, 229, 230, 269, 283, 285, 309, 403-17,
418, 428, 437, 447, 450
additive effects 209, 219, 225, 243, 244,
245
behavior genetics 204, 234, 245, 247, 255,
270, 291
dominance effects 219
molecular genetics 204, 245, 251, 254, 273,
450
quantitative behavior genetics (QBG) 200,
232-33, 240, 242-45, 254, 450
Genghis Khan 269
Genius 342-43, 345, 354
Genome 245, 246-47, 408, 435
Genome-wide association 252, 253
Genotype, genome 19, 25, 205, 214, 242, 243,
252
Gifted (intellectually, class of people) 57, 343,
354/ 384
development of 349
Glenn, John (US Senator, Astronaut) 364
Grade point average (GPA) 28, 120, 125, 311,
312, 314, 322-23
high school 320, 327
college/university 424
graduate education 327
Grades, see academic achievement
Graduate Management Admissions Test
(GMAT) 126
Graduate Record Examination (GRE) 328,
398
Graduation
high school 319, 327
college or university 322, 327, 345
graduate program 327, 328
Graduation rate 59, 61
Gray, Hanna Holburn 397
Gray matter 175, 184, 185, 187, 404
Great Depression (1929-1930S) 345
Gross Domestic Product per capita
(GDP/c) 375, 437, 438, 442, 443
Group differences in intelligence 357, 360,
400
estimating size of 361
guidelines for research on 358
statistical issues 361
Guthrie, Arlo 210
Guthrie, Woodrow (Woody) 210, 248
g-VPR model (four-stratum model) 72, 87,
106-109, 110, 389, 391, 403, 405, 406, 449,
451
first stratum in 106-109
focused-diffuse bipolar factor 388
fourth stratum in 107
SUBJECT INDEX
499
P (perceptual] factor in 108, 166, 167, 192,
250, 263, 364, 389
R (rotation] factor in 108, 139, 166, 167, 192,
199, 236, 250, 263, 305, 363, 364, 380, 389,
4°3, 405, 406
second stratum in 106
third stratum in 106
Verbal (V] factor in 24, 108, 192, 263, 406
verbal-rotational bipolar factor 388, 450
H. M. (Henry Moulson] 182, 184; 196
Habituation 198, 272, 276
Haploid 247, 252
Harding, Warren (US President] 2, 185
Hardy -Weinberg equilibrium 207
Harrison, Benjamin (US President] 218
Harrison, William (US President] 218
Hatshepsut (Pharoah] 376
HBB gene 212
Head Start program 69, 288-91, 295,
310
Health 16, 22, 296, 410, 436, 437, 439
prenatal and infant 271-73
Height 7, 8
Hemisphere (of brain] 174
left 174, 188, 191
right 174, 191, 193
Henry Ford Museum 327
Henry VIII (King of England] 176
Heritability 226-45
broad sense defined 215
coefficient ( h ] 109, 215, 216-17, 225, 227, 230,
232, 236, 244, 254, 261-69, 3°9/ 35^ 434/
443/ 45°
Herodotus 356, 436, 444
Heterozygous, definition of 205
High school 2, 39, 51, 56, 117, 132, 133, 158, 290,
296, 303, 319, 427, 430
drop-outs from 298
High school students 303, 363
Hippocampus 174, 182, 195-98, 199, 202, 270,
4°5
Hispanic, see Latinos
Home (family] environment 284, 287-88, 291,
349/ 427/ 42§
Home Observation for the Measurement of
the Environment (HOME] scale 287,
288
Homeless (category of persons] 343
Homer 1
The Odyssey 1
Homicide rate 442
Homozygous, definition of 205
Honesty 22
Hopper, Grace 397
Hormones 405, 406, 446
estrogens 406
male 405
testosterone 406
Huarte de San Juan, Juan 2, 15, 34
Human capital 436
Humanities 320
Humors (Medieval medical concept] 2
Hunger Winter 276
Huntington’s disease 208, 209, 210, 237,
248-49, 255
Hypothalamus 175
Idiot savant 116
Illiteracy 13
Imagery, see visual imagery
Imagination 2
Imaging (of brain) 304
functional magnetic resonance imaging
(fMRI) 70, 75
X-ray 75
Immigrant 413, 415, 425, 426, 427, 431
Income (earnings, wealth) 22, 260, 266, 284,
287, 311, 312, 319, 326, 327, 336, 337, 341,
342/ 35°/ 354/ 410/ 420, 428, 432, 436/ 437/
439/ 443/ 45°
Indian (Asian or Asian-derived group) 412
Industrial 17
Infant 198
Information processing 21, 73-75 , 76-77, 84,
91, 110, 142, 158, 162, 164, 169-71, 173, 188,
200, 238-39, 272, 277, 301, 371, 445
training of 304, 306
Inspection time 151-52, 153, 239
Integrity, test of 330
Intellectual engagement (personality
variable) 134-35, 258/ 31Q/ 375/ 45°
Intelligence 7, 19, 21, 204, 206, 208, 209, 240,
243/ 247/ 25h 270, 34h 35°
as ability to learn 4
as “can do” concept 357
as knowledge (PPIK model) 132
as part of a system of variables 27, 28,
160
as process (PPIK model) 131
as set of skills 4, 23, 426, 437, 448
belief about 355
biological explanation 2
causes of 19, 24, 30, 260
children’s 220
coaching programs for 301-302
development of 19, 27, 385, 430
effect of lead on 450
gene for 254
group differences in 355
500
SUBJECT INDEX
Intelligence (conr.)
national 436, 450; problems in
estimating 437
polygenetic model of 213
practical 60
Lippman-Boring debate on definition 10, 44
nature of 1, 407
scores on test of 2, 15, 16, 22, 23, 28, 154, 231,
399
stratification of society by 355
training of 300-307
uses of 22
Intelligence Quotient (IQ) 5, 7, 9, 11, 13, 18,
20, 21, 23, 25, 27, 30, 53-54, 57, 68, 98, 114,
122, 132, 136, 151, 154, 182, 184-85, 187, 188,
198, 201, 204, 213-15, 220, 222, 230, 233, 247,
250, 251, 253, 264, 265, 267, 270, 272, 279,
281, 285, 286, 289-90, 299, 300, 310, 317,
319, 332, 334, 336, 337, 341, 343, 345, 349,
377/ 378/ 38l< 384/ 385/ 4°4/ 4°7. 4*3» 4“/
423, 424, 432, 433, 444, 445
defined 6
international values compared 438-40,
441
Intelligence testing 14, 16
Intelligent Design 64
Interests 330, 402
International comparisons of
intelligence 436-45
International Mathematics and Science
Studies (TIMSS) 421
Interview
structured 332
unstructured 331
Inuit 127
Iraq 377, 378-82
Iron 277
Item independence 60
Item response theory (IRT) 52, 55, 57, 363,
368, 369
Item selection 51, 52
Japanese-Americans 421
Jefferson, Thomas (US President) 2, 31, 185,
436
Jews, Jewish 414
Jindal, Bobby (Governor, US State) 431
Job performance (workplace
performance) 16, 68, 91, 119, 122, 135, 324,
329, 330, 337, 350, 416, 420-21, 426
Job requirements 339
Job status 342
Johnson, Lyndon (US President) 172
Johnson, Tim (US Senator) 140
Judgment of motion 389, 403
K-12 school system 312, 319-21, 322, 324,
392-93, 417
drop-outs from 312
Kaufmann intelligence test (KAIT) 34-36, 91,
100, 306
Kennedy, Edward (US Senator) 28
Kennedy, John (US President) 28
Kennedy, Joseph 28
Kennedy, Robert (US Senator) 28
Kindergarten 117, 430
Kleinefelter's syndrome (XX Y) 251
Knowledge 20, 96, 100, 108, 127, 131, 132, 133,
139, 171, 268, 297
academic 267
culture-specific 131, 444
domains of, in PPIK model 132
job-relevant (including military duties) 329,
33h 332
specialized (Gk) 131-33, 139
Korsakoff’s syndrome 278
Kuru 275
Language 21, 27, 47-48, 73, 115, 193, 251, 321,
39h 392
Larry P. v. Riles (court case) 422, 423
Latent variables 18, 19, 158
Latin-American 18, 439
Latinos (ethnic group) 345, 348, 408, 410, 411,
412, 413, 417, 420, 423, 424, 425, 426, 427,
428, 429, 431, 432, 434
definition of 411
Law School Entrance Examination
(LSAT) 328
Lead 280, 432
as atmospheric pollutant, see pollutants
blood concentration levels 282
contained in toys 280
use in gasoline 280
Leadership 126, 127
Learning (practice) 200, 242, 401, 402, 403,
406, 427
Lecithin 257
Lewinsky, Monica 95
Lexical identification 147-51, 153, 155, 161
Lexical retrieval 99
Life, length (life expectancy) 14, 438
Limbic system 137, 174, 175, 182, 188, 196,
278
Lincoln, Abraham (US President) 1, 61, 131
Linguistic performance 242
Linkage analysis 252, 283
Literacy, see reading
Literacy rate 437, 438, 440
Living skills, changes in, with age 374
Locke, Gary (Governor, US State) 431
SUBJECT INDEX
501
Logic 308
Longitudinal study 133, 261, 366, 375
Low-normal intelligence 349-53, 354
Madoff, Bernard 300
Magnetic resonance imaging (MRI) 177, 179,
180, 184
Malaria 212
Male/female ratios
in SMPY 348, 384
in special education programs 384
in student bodies 327, 381
Male-female differences (sex differences) 48,
109, 174, 194, 355, 357, 359, 387, 388
in brain organization 404, 405
in education, educational topics 391, 393,
394“97
in general intelligence 375-82, 384, 386, 404,
4°7, 446
in high-level test scorers 385, 393
in variability of test scores 382-83, 395
international comparisons 393, 401
Male-female differences, causes of 398-406
biological 402
social 400-402
Malnutrition 274, 432
long-term 276
short-term 275, 276
Managerial/professional positions (reference
executive, professional) 332-34
Manic-depressive psychosis (bipolar
disorder) 343
Manifest variables 18, 19
Map reading 390
Marianas Islands 124
Masai (African tribal group) 377
Master of Business Administration (MBA) 126
Maternal intelligence, see parents, parental
Mathematical skill 106, 240, 242, 252, 347, 391,
392, 446
Mathematicians 305
Mathematics 240, 241-397, 398, 400, 401, 419,
421, 426, 428, 442
Mating practices, see assortative mating
Maturation rate, sex differences in 378-413,
4U
McCain, John (US Senator) 162, 165
McNamara, Robert (US Secretary of
Defense) 350
Measurement invariance 404
Medicine 20
Memory 2, 98, 106, 174
conscious 104-105
content 236
declarative 182, 196
episodic 196, 199
for content of written passage 388
for shapes (MS factor) 166
immediate (short-term) 91, 96
implicit 196
long-term 77, 105, 143-52, 165, 196, 249, 305,
367> 374
procedural 196
semantic 196
short-term (storage) 96, 99, 132, 144, 147,
149, 155, 165, 210, 239, 249, 306, 367, 372,
406, 415
unconscious 105
working 73-75, 77, 84, 105, 110, 144, 145, 147,
155-60, 163-64, 170, 175, 189, 190^2, 200,
238-39, 270, 271, 272, 276, 277, 278, 304,
3°8, 353. 37°. 372, 374. 433. 449. 45°;
training of 306-307
Memory scale (on Wechsler tests) 34
Memory span task 74, 157, 158
reading span 74-75
Men (boys) (in context of specific
male-female comparisons) 336, 337, 345,
348, 350, 355, 356, 359, 362-63, 364, 380,
387, 388, 389, 391, 395, 396, 403, 404, 426,
446
eminent 375
Mendel, Gregor 204-205, 206, 209, 210, 283
MENSA 8
Mental calculation (arithmetical) 305
Mental disability or retardation 13, 57, 153,
161, 204, 247, 248, 250, 251, 255, 276, 342,
343. 385. 422, 439
Mestizos (Latin American group) 412
Meta-analysis 318, 323
Metabolism 190, 198
Meta-inventions 283, 309
Mexico 412, 413
MICROCEPHALIN gene 252
Middle school 28, 92, 103, 117, 132, 319
Military 2, 13, 20, 28, 39, 139, 293, 299, 333, 350,
351
Military enlistees 157, 276, 319, 411
Vietnam War veterans 414
Military performance, dimensions of
discipline 329
fitness and bearing 329
general military proficiency 329
leadership 329
technical proficiency 329
Military performance, general 350, 378
Miller Analogies Test 33, 333
Milne, A. A. 183
Mind-set 136
Mini Mental Status Examination 33
502
SUBJECT INDEX
Minnesota Study of Twins Raised Apart
(MISTRA) 107, 233-35, 236- 387, 388
Minnesota Trans-racial adoption study
(MTRA) 231, 285
Minority groups 18
Mislevy, Robert 12
Mitochondria 246
Mixed-race groups 412, 434, 439, 443
Moray House Test 14
Mortality 15
Motivation 30, 139, 396, 402, 426
Motor processes 151
Motor skills 182
Movement time 150, 153
Mozart effect 274
Multidimensional Aptitude Battery (MAB) 91
Multigroup confirmatory factor analysis 364,
365, 378, 413, 415, 426
Multiple intelligences, theory of 114
bodily /kinesthetic intelligence 114, 115, 116,
449
interpersonal intelligence 116, 118
intrapersonal intelligence 116
linguistic intelligence 114, 115, 116, 118
logico-mathematical intelligence 116
musical intelligence 114
naturalistic intelligence 116
spatial intelligence 116
Mutualism 96
Myelin 175
Name identification 148
Napoleon (Napoleon Bonaparte) 297
National Assessment of Educational Progress
(NAEP) 267, 383, 392, 403-17, 418, 421
National Association for the Advancement of
Colored People (NAACP) 411
National cognitive skill
as a general factor in international
scores 442
index of, related to other variables 442-43
National Council of Educational Statistics 39
National differences in intelligence 357, 359
National Education Association (NEA) 39
National Football League 46
National Longitudinal Study of Youth
(NLSY) 292, 318, 339
NLSY79 292, 293-94, 299, 319, 320, 323, 336,
350, 379, 420
NLSY97 292/ 322/ 379/ 387
NLSY Children and Young Adults
Study 292
Native American, see Amerindian
Natural kind 104, 116
Nature-nurture debate 66-68
n-back task 306
Netherlands, the 414
Neural activity 176
Neural conduction 109
Neural density 199
Neural efficiency 254
Neural plasticity 109, 198-99
Neurons 175
Neuropsychology 176
Neuroscience 77
New Guinea 275
Newton, Isaac 65, 451
No Child Left Behind Act 313
Normal (Gaussian or Bell) distribution 6, 7
Norming 56
Nostradamus 65
Number (numerical) facility 98, 99, 336, 340,
341
representation in brain 305
Nutrition 260, 270, 273-77, 283/ 2&7> 29^/ 3°7/
309, 434, 439
0*NET 340
Obama, Barack (US President) 280, 445
Occipital cortex 37, 174, 180, 181, 193, 199
Occupation 284, 287, 338, 428, 450
blue-collar 351
classes of 337, 340
managerial/professional 351
status 339
white-collar 351
Odds ratio 328
Officers Qualifying Test (OQT) 23
Old age 9
Orientation, orienting ability
(navigation) 166, 167-69, 195, 199, 200,
403, 405, 407, 443
Otis-Lennon Test of School Ability 40-41, 320
Panel Study on Income Dynamics (PSID) 428
Parameters
fixed, definition of 222
free, definition of 222
Parenting practice 260, 287, 294, 428, 432, 435
Parents, parental 283, 284
adopting 230, 231
biological 229, 230, 231
intelligence 69, 270, 272, 279-81, 285, 287,
292-94
Parietal cortex 37, 174, 180, 181, 184, 185, 188,
190, 193, 195, 198, 199, 200, 202, 306, 372,
373/ 449
Parieto-frontal integration theory (P-FIT) of
brain function 190, 191, 202
Parkinson's disease 271, 375
SUBJECT INDEX
5°3
Pascal, Blaise 27
Path analysis 221
Path coefficients, definition of 221
Pattern recognition 109
Pedigree study, definition of 205
Peers, peer group 308, 447
Penetrance, definition of 216
Perceptual skill 72, 106
Perceptual: motor factor 106, 334, 341-52
Performance IQ (PIQ) 31, 34, 188
Personality 22, 30, 122, 132, 134-37, 139, 329,
33°; 354; 45°
Personnel selection and classification 13, 16,
18, 48, 56-59, 77, 110, 324, 331, 380
in college entrance programs 402
racial/ethnic discrimination in 425
sex discrimination in 380
Phenotype, definition of 205, 219
Phenylketonuria (PKU) 248, 250, 255
Phi Beta Kappa 12
Phrenology 186
Physical fitness 10
Physical identification 148
Physics, teaching of 303-304
Picasso, Pablo 61, 343
Planning, Action, Simultaneous, Serial
(PASS) Model 37, 449
Policy issues 62, 291, 296, 358, 409, 416, 417,
425; 447
Political development (of country) 443
Pollutants 270
Kosovo study 281, 288
lead 279-309
mercury 279
Port Pirie study 281
Polygenetic model 212, 251
Polymorphisms, defined 253
Polynesian navigators 297
Positive manifold 80, 87, 91-94, 97, 109, 110-15
Positron emission tomography (PET) 178,
179, 180, 189
Postgraduate education 327-28, 333, 397
Post-secondary education (college, university,
graduate) 297
Post-secondary system of education 322
Poverty 427
Power (of statistical tests) 315, 317, 361, 379
Practical intelligence 121, 122, 123, 125, 126, 128,
129; *3°; 301; 332; 373
Practice 154, 400, 446
Precocity
mathematical 346
verbal 346
Predictive validity 330, 331, 354
Premature birth 272
Pre-school programs 69, 274, 296
Presidency, American 218, 343
Prestige 336, 337
Primary abilities model of intelligence 97,
107, 264
Prison (jail) 349
Problem solving 2, 393
Procedural knowledge 122
Processes, personality, interests, and
knowledge (PPIK) model 130-34
Program for International Student Assessment
(PISA) 392, 393, 394, 401, 421, 437, 441
Progressive Matrix item format 46, 47, 48, 50,
51, 52, 96, 160, 161, 188, 191, 266, 267, 297,
302, 303, 329, 380, 396, 397, 405, 425
Project 100, 350, 351, 352, 353
Prospective study 318, 334, 336, 343, 350
Protein 277
Proximal influences 68, 215
Psychometrics 11, 52, 72-73, 76, 77, 79-80, 91,
120, 130, 138, 142, 143, 153, 166, 199, 238,
239; 242, 367, 371, 386, 389
Quality of Human Conditions (QHC)
index 438
Quasi-experimental design 360
Quiz Kids 344
Race track handicapper study 317
Racial/ethnic differences ‘gap” 235, 417, 443,
447; 45°
biological/environmental causes for 432
genetic causes for 433-35, 443
in educational achievement 415-20
in intelligence and test scores 256, 355, 357,
359; 404, 407, 411-13, 414-15, 433
Jensen-Rushton “default hypothesis”
for 434-35
social causes for 425-32
Racial/ethnic groups 18, 255, 356, 408, 409,
439; 443; 446; 447; 45°
RAINBOW study 125, 126
Ramanujam 354
Range restriction, effect of 314, 315, 316, 320,
321; 33°
Ratings
supervisor ratings of employee
performance 312, 420
teachers’ ratings of students 240, 241
Raven Progressive Matrices Test (RPM) 46,
51, 56, 91, 95, 96, 190, 302, 330, 378,
380-82, 411, 413, 427
Advanced 33, 46, 189, 190, 284, 303, 322, 333,
3Sl
Colored 46
SUBJECT INDEX
5°4
Raven Progressive Matrices Test (RPM) [ cont .)
Standard 46, 266, 268, 303, 321, 413
standardization study of 268
Reaction range 24, 25, 26, 37, 258, 286, 307, 437
Reaction time 154, 372, 433
Reading 20, 27, 47, 68, 132, 193, 240, 241, 442
Reagan, Ronald (US President) 2, 364
Reasoning, 40, 100, 132, 145, 158, 188, 190, 199,
200, 263, 297, 300, 302, 303, 304, 334, 335,
34°, 341, 360, 391, 415
abstract 2, 50, 115, 267, 272, 278, 387
American vs. Asian style of 50
auditory 367
counterfactual 431
deductive 34—38, 49-50, 99
inductive 34, 46, 49-50, 95, 98, 99, 263, 300,
369
logical 161
mathematical 48, 143, 161, 192, 283, 357-59,
399/ 431
nonverbal 271
quantitative 94, 95
scientific 308, 399
verbal [see also verbal ability) 96, 127, 143,
H5> l89« *99< 200< 25°/ 278. 34°< 37°/ 387>
4H
visual-spatial 49, 145, 174, 189, 192, 199, 200,
263, 278, 308, 367, 370, 380, 387, 389, 401,
406, 414, 421, 446
Recessive trait, definition of 205
Recommendations 3
Recruitment effect 359, 380, 410, 411, 412, 437,
439
Reductionism 71-72
Rejection rate (in personnel selection) 325, 331
Reliability (statistical concept) 154, 188,
313-14, 316, 317, 318, 330
Replacement rate 208
Research design 318
Retrospective study 319, 343
Rice, Condoleeza 28
Robert Bruce (King of Scotland) 184-85
Roberts, John (US Chief Justice) 283
Robin Hood 300
Rodriguez, Alex 283
Roles in society, men compared to
women 377, 400, 401, 402, 446
Roma (European ethnic group) 412
Roosevelt, Franklin (US President) 39, 131, 218
Roosevelt, Theodore (US President) 218, 298,
300, 326
Rotation (of visual images) 72, 106-109, *69,
3°5> 367« 389. 39°
Rousseau, Jean-Jacque 61
Russia 127, 131
Saint Paul (Saul of Tarsus) 65
Salem witch trials 275
SAT 9, 14, 15, 17, 21, 23, 32-33, 39, 41-43, 48,
50-51/ 54. 56/ 58- 59/ 61, 62, 73/ 103-104,
120, 121, I25, 126, 135, l6o, 164, 217, 219, 265,
z84/ 297/ 3°i/ 302/ 313/ 322_24/ 325/ 326~27/
333/ 346/ 359/ 36°/ 385/ 395~97/ 418/ 419.
421, 422, 424
Saturated model, definition of 223
Schizophrenia 343
Scholastic ability 40
Scholastic Aptitute Test, see SAT
School, schooling 28, 114, 117-18, 277, 447
Science, knowledge of 241-398
Science, Technology, Engineering,
Mathematics (STEM) 392, 395-98, 399,
400, 402
Seattle Longitudinal Study 263, 265, 268, 318,
367, 369, 370, 371, 373, 375, 451
Segregation (of racial-ethnic groups) 355
Selection pressure, 208, 210, 416, 445
driven by climatic conditions 444
Selection ratio (complement of rejection
rate) 331
Selection restriction (in personnel
classification) 314, 315
Self-description 137
Self-discipline (personality variable) 135-36
Semantic identification 147
Semantic priming 163
Sensory processes 151
Sentence comprehension 99, 141, 142
Sentence verification paradigm 195
Serb (European ethnic group) 412
Serial processing 147
Sesame Street (TV program) 283
Sex chromosomes, definition of 211
Sherlock Holmes 49, 238
Shumer, Charles (US Senator) 162
Siblings (SS) 220, 224-25, 228, 238, 243
Sickle-cell anemia 212
Simpler reaction time 371
Single nucleotide polymorphism (SNP) 247,
253
Sioux (Lakota) 3
Sitting Bull 3
Situation awareness 156
Situation model 74, 161-62, 165, 169
Situational judgment test 333
SNAP-25 gene 251-52
Social achievement 21
Social beliefs, see policy issues
Social complexity, changes in 308, 310
Social drinking 278-79
Social relevance 22
SUBJECT INDEX
5°5
Socioeconomic status (SES) 15, 16, 17, 25, 62,
68, 97, 217, 219, 220, 226, 230, 231, 232, 233,
243, 244/ 255« 258-6o> 27°> 279- 2§3< 284-
286, 287, 288, 29O, 29I, 294, 295, 298, 3O7,
318, 319, 326-27, 337, 345, 349, 360, 414,
421, 427-28, 430, 432
Soldiers (military personnel) 262, 268, 329-30
South Africa (country) 412
Span tasks [see also digit span) 156-57
memory 157
processing 157
reading 163, 189, 239
Spatial orientation 21, 263, 390
Spatial skills 340-50, 403, 415, 446
Spearman's hypothesis 363-64, 365, 414-15
Special education 242, 384
Speed 142, 146, 200, 304
cognitive processing (Gs) 99, 152, 155, 304,
305, 371-72, 373, 374, 387, 445, 450
decision (Gt) 99
information-processing 77, 96, 145-47, 154>
155, 170, 189, 238, 433, 449
of closure (CS) 166
of rotation (SR) 166
perceptual 98, 240, 371, 386
Speed-accuracy tradeoff 151
Spies 4
Standard score 54
definition of 6
Standardization 11, 269
Stanford, Allen 300
Stanford-Binet 7, 12, 13, 32, 34, 40, 122, 265,
269
Stanine, defined 383
Stereotype threat 357-59, 361, 402, 426
Structural equation modeling 158, 221, 237,
244
Structure of intelligence model 106
Study of Mathematically Precocious Youth
(SMPY) 346-49, 354, 366, 374, 398, 399,
400
Styles (of thought) 129
Successful intelligence, theory of 120-30, 449,
451
Sudan 28, 302, 303
Summers, Lawrence 381-400, 415
Supervisor ratings 122
Syntax 75, 141
System (of variables) 28
closed 28
open 28
social 70
variables external to (external or exogenous
variables) 28, 29
variables in (system variables) 28, 29
System analysis 30, 68
Tacit knowledge 122-24, l27> 129> ^8
Taxicab drivers 195, 202
Temporal cortex 37, 75, 174, 180, 181, 193, 197,
198, 199, 200, 258, 270
Terman study of gifted ('Termites") 344-46,
349, 354. 366
Test sophistication, test-taking skills 266, 302
assessed irrelevant skills 423
assessed relevant skills 423
unassessed relevant skills 423, 426
Tests ( see also names of individual tests) 3
appropriateness across racial/ethnic
groups 361
batteries of subtests 31, 40-44, 302, 321, 377,
380, 407, 411
British Empire's use 4
Chinese Empire's use 3
cognitive 2, 8, 13
g-loaded 357
gender bias on 378
group 13, 31-38, 39-47
individual 31, 32, 33-39
information-processing 272
intelligence (IQ) 3, 5, 8, 9, 12, 21, 23, 33,
121-22, 266, 271, 284, 304, 308, 373, 446
mathematics 387
objections to 3, 17, 120-21
performance on 72, 241
scores 133, 137, 154, 189, 201, 204, 220, 241,
242, 254, 260, 261, 265, 266, 270, 293, 295,
298, 301, 309, 319, 326, 334, 337, 339, 349,
360, 407, 409, 421, 426, 431, 433
single-format 32, 44
situational judgment 122, 126, 129
specific factor evaluated by test 407
validity of 18, 128, 438
verbal 272, 380, 421
visual-spatial 380
TETRIS (computer game) 198
Texas adoption project 230
Text model 73, 161, 169
Theory 64-78
competition between 66-68
scientific 65, 77
Three-minute reasoning test 44, 146
Three-stratum model 99-106, 109, 138, 139,
152, 278, 367, 370, 371, 391, 449, 451
first stratum 99
second stratum 99, 100
processing speed (PS) factor 389, 449
third stratum [see also general
intelligence, g) 100
visualization (Vz) factor 389
SUBJECT INDEX
506
trait complex 133
conventional complex 134
intellectual complex 134
science and math complex 133
social complex 134
Truman, Harry (US President) 1, 2
Tuddenham Study 262, 268
Turner’s syndrome 250
Twin studies 189, 232-35, 243
Australian 238, 239
Dutch 233, 236, 238, 239, 253
Japanese 238
Swedish 233, 235-36, 237, 240, 375
Twins Early Development Study
(TEDS) 240, 241, 242
US Department of Veterans Affairs 239
Twins 243
adopted apart 234, 236
dyzygotic (fraternal) 67, 220, 221, 224-25,
226, 228, 232, 233, 236, 237, 243
monozygotic (identical) (MZ) 25, 67, 220,
221, 226, 228, 232, 234, 236, 237, 243, 272
virtual 228
University of California 62
University of Michigan 50
Validity 14, 18, 21, 58, 63, 426
across racial/ethnic groups 358, 422-25
internationally 439
predictive, defined 312
Variability in test scores, male-female
differences 382-85, 450
Variance ratio, defined 382
Verbal ability (intelligence, skills) 21, 72, 75,
94, 95, 136, 151, 155, 192-93, 268, 378, 403,
446
Verbal ability (linguistic skills) 21, 72, 75, 106,
162, 334, 341, 347, 387, 428
Verbal comprehension 73-74, 100, 106, 146,
I56, l60-66, 169, I76, l80, 189, 263, 3O5,
369, 4O7
Verbal fluency 406
Verbal IQ (VIQ) 31, 34, 188
Verbal relations (primary ability) 98
Verbal: educational factor 106
Verbalization 96
Video games 308
Virtual environment 168, 390
Vision 37
Visual analysis of static figures 390
Visual detection 304
Visual imagery 166, 167, 168, 185, 193, 387
Visual recognition 96
Visualization 95
Visualization factor (Vz) 166
Visual-spatial ability 32, 48, 60, 98, 136,
166-70, 189, 193, 194-95, 382, 397, 400,
426
Gv 99, 102, 103, 136, 166, 167, 168, 194, 199
training of 389
Vitruvius 280
Vocabulary, vocabulary tests 1, 2, 32, 33, 91,
106, 118, 129, 163, 169, 193, 200, 258, 272,
302, 367
Washington, George (US President) 20, 131,
?>l9> 343
Watson, James 344, 416-17
Wechsler Adult Intelligence Test (WAIS) 12,
U, 34' 35' 3b' 39/ 46, 91, 92' 94' 98, 103; 155,
163, 182, 188, 193, 200, 201, 230, 236, 265,
284, 3*7. 354. 363-75. 378, 379. 38o, 383,
386, 387, 407
Wechsler Intelligence Scale for Children
(WISC) 12, 34, 57, 98, 232, 279, 286, 288,
380
Wechsler Preschool Intelligence Test 290
Wechsler tests 12, 13, 31, 32, 40, 56, 269, 319,
377. 378
Weight 8
Welfare (public assistance) 349, 354, 359
Wernicke’s area 75, 175, 193
White (racial-ethnic group) 18, 226, 231, 232,
235, 261, 272, 283, 290, 336, 337, 339, 345,
35°. 361. 363, 364, 4°4. 4°8, 4°9. 410.
411-13, 415-19, 420, 421-22, 423-24, 425, 427,
428-30, 432, 433, 434, 443, 446
White matter 175, 187, 199, 404
Williams syndrome 161
Wilson, Woodrow (US President) 2
Wisdom 24, 373, 374
Wollstonecraft, Mary 376
A Vindication of the Rights of Women 376
Women (girls) (in context of specific
male:female comparisons) 305, 334, 336,
337. 345. 348, 350. 355. 35b. 359. 362, 364.
380, 386, 387, 388, 389, 391, 395, 396, 403,
4°4. 42b. 44b
eminent 375, 397
Wonderlic Personnel Test (WPT) 44-46, 48,
73, 118, 171, 330, 338, 420
Woodcock-Johnson Intelligence Test
(WJT) 34, 36, 100-103, 367, 368, 369, 370,
371' 412' 4 U, 4^
Word fluency (primary ability) 98, 106
Word (lexical) recognition 163
Work sample 331
SUBJECT INDEX
507
Workplace, cognitive requirements of 350
Workplace performance 328-50
Workplace rewards (for intelligence) 334
World War I 10, 261, 262, 263, 280, 411,
448
World War II 235, 261, 262, 263, 265, 267, 275,
276, 296, 298, 322, 334, 354, 411, 440
Wundt, Wilhelm 89
X-rays 176, 177
XYY syndrome 250-51
Yeltsin, Boris (Russian President) 131
Yerkes, Robert 411
“Earl Hunt has written a book that is deeply informed by a scholarly knowledge of
the literature. No one in the field has written in a less partisan and less polemical
manner. What impressed me greatly about the book is the extraordinary level of
pedagogical skill that is apparent in his writing. He has an unusual ability to focus on
the most important studies in the field and to summarize their essential features in a
clear and compelling manner. This book will be accessible to many individuals with
diverse backgrounds.”
- Nathan Brody, Wesleyan University
“Hunt reviews the main topics in the field, from socially relevant and measurement-
related affairs to brain research, the interplay between genes and environment, and
e group differences in intelligence (age, sex, and race/ethnicity),
ssions follow a clear presentation based on a revision of the available
evidence. The resulting text is vibrant, brilliant, and inspiring,
gnificant contribution to future generations. A new wave of
land on the shoulders of this giant book.”
erto Colom, Universidad Autonoma de Madrid
“An up-to-date, welt-n inded, thoughtful, objective, readable book on one of the most
interesting areas of ps> :hology.”
- Ian J. Deary, The University of Edinburgh
Earl Hunt is Professor Emeritus at the University of Washington, where he has been a
faculty member since 1966.
Cambridge
UNIVERSITY PRESS
www.cambridge.org
Cover image: “Intelligence.” Courtesy of iStockphoto
Cover design by fred goykhman/PETT Fox, Inc.
Live Music
Archive
Librivox
Free Audio
Metropolitan Museum
Cleveland
Museum of Art
Internet
Arcade
Console Living Room
Open Library
American
Libraries
TV News
Understanding
9/11